In October 2026, two of the biggest SEO tool companies said the same thing in different words. Semrush published a guide to running an SEO audit by asking Claude questions. Ahrefs published a walkthrough for automating content updates with an AI agent. And a Search Engine Land contributor predicted that by 2027, paid-search agencies will mostly direct automated systems rather than do the work by hand.
If you pay someone for SEO every month, the obvious question follows: if AI can do SEO now, what exactly am I paying for? This guide gives you a straight answer, a 10-minute test you can run on your own site, and five questions to put to any SEO provider, including us.
Quick answer
Can AI do SEO? Partly. AI can now do much of the mechanical work: reading audits, sorting issues, drafting titles and first-draft content, and explaining reports. It cannot fix your website, decide what matters to your business, or take responsibility for results. Without real data, it will sometimes report an audit it never ran.
Pay less for
Audit exports, keyword lists, generic blog posts and reports that only restate numbers.
Keep paying for
Judgment about what matters, hands-on website fixes, content with real experience, and accountability.
What Changed in October 2026
The news is not that AI can talk about SEO. It is that AI assistants now plug straight into the data SEO tools collect, through connections called MCP servers.
Semrush: Claude reads real audit data
Semrush's guide to auditing with Claude (15 October 2026) shows Claude reading completed Site Audit results: health scores, errors versus warnings, affected pages, likely causes, priorities and which team should own each fix. It can also draft replacement meta descriptions and title tags. Semrush is clear about the limits. Claude can only read audits that have already finished. It cannot start a crawl or change settings, and each request spends between 100 and 10,000 paid API units.
Ahrefs: automation, with a human sign-off
Si Quan Ong's follow-along guide to automating content (16 October 2026) describes a system that refreshes 12 data-driven articles every month. It pulls fresh data, filters out junk, rebuilds tables and charts, and updates WordPress drafts. Then it waits. His one fixed rule is that AI never publishes on its own, because someone has to be responsible for what goes live. His summary of the division of labour is worth remembering: the LLM's job is "to process the data, not to be the data."
Two of the biggest SEO tool vendors, one message. AI handles the repeatable parts. People own the data sources, the judgment and the sign-off.
What AI Can and Cannot Do in SEO Today
AI is strong at reading, sorting and drafting. It is weak at deciding, doing and owning. Here is the realistic split for a typical small or mid-sized business website.
SEO task
Can AI do it?
The catch
Summarising a technical audit
Yes, well
Only when it reads a real crawl from a real tool
Writing title tags and meta descriptions
Yes
Needs a human check for accuracy and brand voice
Keyword lists and topic ideas
Yes
Volumes are estimates; it does not know your margins
Monthly reporting
Mostly
It explains numbers, but not which ones matter to your business
First drafts of content
Yes
Generic unless someone adds real experience and facts
Deciding what to prioritise
Partly
It ranks issues by severity, not by what makes you money
Fixing the website
Not safely alone
Templates, speed, redirects and hosting need someone accountable
Being accountable for results
No
A chatbot does not answer for a bad quarter
A note on fixing the website: connected agents can now edit pages through tools such as the WordPress API. That makes human review more important, not less. An unchecked bulk change can break a site faster than any person could.
The Catch: AI Will Sometimes Invent the Audit
This is the part most AI SEO tool sellers leave out.
What a 2025 study found when ChatGPT audited real websites
In a peer-reviewed 2025 study, Karol Król of the University of Agriculture in Krakow built two ChatGPT-based website auditors and tested them on two of his own WordPress sites against standard tools such as PageSpeed Insights and a broken-link checker. The results were confident, well formatted and often wrong.
Check
AI auditor reported
Real tools measured
HTTP requests on the page
45
133
Total page size
1.2 MB
10 MB
XML sitemap
Missing
Present at the standard address
Link audit
Errors flagged, fixes suggested
No link check was ever run
Source: Król, K. (2025), Applied Sciences 15(5), 2292. Results for the first test site, desktop mode.
Each time, the model claimed the audit had been performed. Only after repeated, detailed questions did it concede the figures were illustrations of what a real tool might show. Król's conclusion: using AI tools without understanding how they work can carry hallucinations straight into audit reports.
Why AI connected to real SEO data is different
AI has improved since that study, and connected assistants that browse or read tool data behave better. But the lesson holds. An AI SEO answer is only as good as the data it actually read. A Semrush-connected Claude reading a finished crawl is working from evidence. A plain chatbot asked to "audit my site" may be describing a typical website, not yours, and it rarely says so unless you ask.
Example: Two Audits of One Website
This is an illustrative example built from patterns common on small-business sites, not a specific client.
Picture a plumbing company with 40 pages. Enquiries have dropped since the site was redesigned three months ago. The owner asks a chatbot for a full SEO audit, then has a person run a crawl and open Search Console.
Plain chatbot "audit"
Real crawl + Search Console
Top issue
Meta descriptions could be more compelling
Old service URLs redirect in chains of 3 hops after the redesign
Second issue
Add more location keywords
The emergency plumbing page is "Crawled, currently not indexed"
Third issue
Start a blog to build authority
Mobile layout pushes the phone number below the fold
Where it came from
Not stated; generic advice
Named tool, named report, named URLs
Link to lost enquiries
None
Direct: the money page dropped out of Google
What the AI audit missed, and why it mattered
The chatbot's list was not wrong, exactly. Better meta descriptions rarely hurt. But it could not see the redirect chains or the de-indexed page, so it missed the reason enquiries fell. That gap, between advice that sounds right and evidence about your site, is where most of the value in SEO now sits.
The fix itself is the second half. Someone has to clean up the redirects, find out why Google dropped the page, change the template and then check that enquiries recover. AI can help at every step. It cannot own the outcome.
A 10-Minute AI SEO Audit Test You Can Run Yourself
You can see the difference for yourself before you spend anything.
The test · 10 minutes
1
Ask for an audit
Open any AI chatbot and ask: "Do a full SEO audit of [your website] and list the top 10 problems."
2
Ask for its evidence
Then ask: "Which exact pages did you fetch, and which tool measured each problem?"
3
Open real data
Open Google Search Console (free) and look at the Pages report and the Core Web Vitals report for the same site.
4
Compare the two lists
Check which of the AI's problems appear in real data, and which real problems the AI missed.
How to score the AI's answers
Mark each of the AI's problems as one of three things:
Verified
It shows up in Search Console or on the page itself.
Generic
Advice that would apply to any website.
Unverifiable
The AI cannot say where it came from.
Then check the other direction. Search Console shows things a chatbot has no way to see, such as pages Google has chosen not to index. If those are missing from the AI's list, the AI was not auditing your site. It was describing a typical one.
Finally, look at the last report your SEO provider sent you. If it is a tool export with a logo on top and no explanation of what it means for your business, that tells you something too.
What Google Says About AI-Made SEO Work
Google does not care who typed the words. It cares whether the result is accurate and useful.
Its guidance on generative AI content says AI can be particularly useful for researching a topic and adding structure to original content. The same page warns that using AI to generate many pages without adding value for users may breach its spam policy on scaled content abuse.
On 1 October 2026, Google rewrote the accuracy section of that page. It now says a human check of AI-written material is critical before it goes live, covering everything from body text to page titles and image descriptions. And the September 2026 spam update listed scaled content abuse among the policies in scope.
In plain terms: AI-assisted work is fine. Unchecked AI work at volume is a risk you are paying for.
What to Stop Paying For, and What Is Still Worth It
The dividing line is simple. If a machine can now produce it in minutes, stop paying monthly for it. If it needs judgment, hands on your website or someone accountable, keep paying for it.
Stop paying for
Raw audit exports. A 40-page PDF of tool warnings now costs almost nothing to produce.
Keyword lists on their own. Any tool generates 500 in seconds. The value is in choosing the five that matter.
Bulk generic blog posts. Ten posts a month with no input from you is exactly what Google's scaled-content policy warns about.
Reports that only restate numbers. "Traffic up 4%" is instant with AI. A report should tell you what to do next.
Still worth paying a person for
Deciding what matters. A strategist ranks issues by revenue: the service page that brings enquiries comes before 300 warnings on old blog tags. That prioritisation is the core of real SEO services.
Actually fixing the site. Templates, redirects, speed and hosting changes need someone who can do them safely, and undo them if needed. That is website development work.
Content with real experience. Your prices, process, customer questions and photos of real jobs. That needs someone who talks to you.
Accountability. When rankings drop after a site change, you need a person who finds out why and answers for it.
If most of your current retainer sits in the first list, you are probably overpaying, whoever you work with.
Five Questions to Ask Your SEO Provider
You do not need to ban AI from your SEO. You need to know where it is used and who checks it.
Question 1Which parts of my work are done by AI, and which by a person?
Question 2Does the AI read real crawl and Search Console data, or does it just answer questions?
Question 3Who reviews AI output before it reaches my website?
Question 4Now that audits and reports are cheaper to produce, what am I getting that I wasn't before?
Question 5What did you change on my site last month, and what happened as a result?
Vague answers to questions 2 and 5 are the warning signs. A good provider will answer all five without hesitation, and will welcome being asked.
Frequently Asked Questions
Can AI do SEO on its own?
Not fully. AI can summarise audits, draft titles and content, and explain reports, especially when connected to real tool data. It cannot reliably fix your website, decide what matters to your business, or tell when its own answer is wrong. Treat it as a fast assistant, not a replacement for someone accountable.
Will AI replace SEO agencies?
It will replace the parts of agency work that were mostly mechanical: tool exports, keyword lists, bulk generic content and number-only reports. Agencies whose value was mainly those things will struggle. Strategy, implementation, real-experience content and accountability remain human work.
Can I use ChatGPT or Claude to audit my website?
Use it to explain an audit, but be careful asking it to perform one. A peer-reviewed 2025 study found ChatGPT-based auditors reported measurements and link checks that never happened. Run a real crawl or check Google Search Console first, then ask AI to help interpret that data.
Does Google penalise SEO work done with AI?
Not for using AI as such. Google says AI can help with research and structure. It warns against generating many pages without adding value, and since October 2026 its guidance calls a human check of AI output critical before publishing.
Should AI make my SEO cheaper?
Some of it should. Audits, reports and first drafts cost less to produce, and you should not pay old prices for them. Judgment, website changes and real-experience content have not become cheaper, and they are where most results come from.
The Takeaway
AI can do the reading, sorting and drafting of SEO faster and cheaper than any person. It cannot decide, fix or answer for the result, and without real data it will sometimes make things up. Stop paying for what a machine now does almost for free. Keep paying for judgment and hands-on work.
Not sure which side of the line your current SEO falls on? Send us your last monthly SEO report. We'll mark which parts AI could handle for you, which parts still need a person, and answer the five questions above for your own site. No pitch unless you ask for one.
Ajay Hooda is the co-founder of Digital Roof, a founder-led digital marketing studio specializing in SEO, website development and content. He holds a PG Diploma in Digital Marketing and Communication from MICA, is Google-certified, and has worked with small businesses and international clients since 2019.
It’s budget time. Many business owners are looking at their marketing costs right now.
They see a line that says “SEO.” Next to it is a traffic chart that is flat or going down. So they ask a simple question: is SEO still worth the money?
It’s a fair question. Even big SEO companies are asking it. In early October, Search Engine Land, Ahrefs, Neil Patel’s team and Semrush all wrote about how hard it has become to prove SEO results.
This guide is for the person who pays for SEO, not the person who does it. You’ll get a simple test to see if SEO makes sense for your business. We’ll also tell you when the honest answer is no.
The quick answer
SEO is worth it if your customers search for what you sell, each customer brings in good money, and you can wait six months or more for results.
SEO is usually not worth it if you need leads next month, sell something nobody searches for, or have a website that doesn’t turn visitors into customers.
Here’s the big change: SEO still works. But the way you measure it must change. If you only look at website traffic, SEO will look worse every month, even when it’s helping your business.
What changed? Fewer people click on websites
Most Google searches now end without a click
A research company called SparkToro studied Google searches in the US from January to April 2026. It found that about 68% of searches ended without a click to any website.
The main reason is AI Overviews. These are the AI-written answers Google shows at the top of the results page. They now show up on more than one in five searches.
When people get their answer right there, they don’t need to visit a website.
A quick note on the numbers
SparkToro used slightly different data in earlier years, so the exact jump isn’t perfect. Also, the study didn’t include the Google mobile app. So the real number is probably even higher.
What this means for you
The same Google ranking now sends fewer visitors than it did two years ago.
So if your SEO report still focuses only on website visits, it’s measuring the part of search that is shrinking.
Less traffic doesn’t mean SEO stopped working
Here’s a common trap. Imagine this:
A customer asks ChatGPT for advice, or reads a Google AI answer.
They see your business name there.
A week later, they type your business name into Google and visit your site.
Your analytics tool will say this person came from a “brand search” or “direct visit.” The SEO work that put your name in front of them gets no credit.
One estimate from a company called Graphite says AI’s real effect may be up to 10 times bigger than what tracking tools show.
The lesson is simple. If you cut every channel that can’t prove itself on a report, you may cut the ones that are actually working.
The 5-question test: is SEO right for your business?
Answer these questions honestly before you start or renew an SEO budget.
Question
If YES
If NO
1. Do people search on Google for what you sell?
Good. There are buyers looking for you.
SEO can’t create demand. Make people aware of your product first.
2. Is one new customer worth a good amount of money?
Good. One sale can pay for months of work.
Cheap, one-time sales rarely pay back the cost.
3. Can you wait 6 to 12 months for results?
Good. SEO grows slowly but keeps building.
Use Google Ads or referrals for quick leads.
4. Does your website already turn visitors into enquiries?
Good. More visitors will mean more leads.
Fix your website first. More traffic to a weak page wastes money.
5. Do you know things your competitors don’t?
Good. Real expertise is what Google and AI tools reward.
Generic content is easy for anyone to make now. It won’t stand out.
How to score it: look at questions 1 to 4. If you said “yes” to three or more, SEO is a smart investment. If you said “yes” to two or fewer, spend your money somewhere else first.
Not sure how your business scores?
Tell us about your business and we’ll run this test with you. We’ll give you a straight answer, even if it’s “not yet.”
We’d tell you to hold off on SEO in these five cases:
1. You need money coming in within 90 days
SEO is too slow for this. Google Ads, referrals and reaching out to customers directly will work faster.
2. Your product is brand new and nobody searches for it
If no one is searching, there’s nothing to rank for. First, spend money on telling people your product exists.
3. Your website doesn’t turn visitors into customers
If you rarely get enquiries now, more visitors won’t fix that. Improving your website will make you more money than better rankings.
4. Your content only answers simple facts
If Google’s AI can answer the question in one line, people won’t click through to your page. Much of that traffic is already gone.
5. Someone promises you a #1 ranking
Google itself says no one can guarantee a #1 ranking. If an agency promises this, walk away.
How long does SEO take?
Many owners give up on SEO because nobody told them how long it really takes. Here’s what Google’s own data shows.
What Google’s data says
In October, Google’s Gary Illyes shared timing data at an event in Barcelona. This was reported by Search Engine Roundtable. Google hasn’t published the slides, so treat these as rough guides, not promises.
Finding a new page: Google usually finds a new page in about 20 hours. Sometimes it takes weeks, or never happens.
Adding it to Google: once found, a page is usually added to Google’s index in about 1.5 hours. If the page is low quality, it can take months, or never happen.
Recovering from a Google update: this usually takes 3 to 6 months. It can take up to a year.
Why a 3-month trial usually “fails”
Getting your page into Google is fast. But getting Google to trust your website enough to rank it high takes much longer.
That’s why short SEO trials almost always look like failures. Review your SEO results at six months, not six weeks.
A free check you can do this week
You only need two things: Google Search Console (a free tool from Google) and your own customers.
Step 1: compare impressions and clicks
Open Search Console and look at the last 16 months. Impressions are how often your site appeared in Google. Clicks are how often people visited.
If impressions are steady or going up but clicks are going down, people still see you. They’re just getting answers without clicking. That’s a measuring problem, not an SEO problem.
Step 2: check searches for your business name
In the Queries report, filter by your brand name. If more people are searching for you by name, they may have found you somewhere else first, like in an AI answer.
Step 3: ask every new customer one question
Add this to your enquiry form and sales calls: “How did you first hear about us?”
Give options like Google, ChatGPT or other AI tools, social media, and word of mouth. Ahrefs suggests this too, because tracking tools often miss AI’s role.
Step 4: count enquiries, not visits
After 60 days, put three numbers side by side:
Search impressions
Searches for your brand name
Customers who say they found you through Google or AI
If all three are flat, your SEO isn’t doing much. If all three are going up, your SEO is working, even if website visits are down.
The mistake most owners make
They see clicks dropping and decide SEO has failed. They never check the impressions line. If impressions are holding up, SEO is working. Google is just sending fewer people onward.
Want us to run this check for you?
We’ll go through your Search Console data and show you what’s really happening with your SEO, in plain language.
Think of your SEO budget as paying for effort and skill, not guaranteed traffic. Here’s a simple way to plan it.
Pay for your website foundation once
A fast, well-built website is a one-time cost. It helps every marketing channel, not just SEO. So it belongs in your website development budget, not a monthly fee.
Pay monthly for real expertise
Your monthly spend should go on content only you can write. Think real prices, real processes and real customer stories. That’s the kind of content creation AI can’t copy.
Agree on three numbers before work starts
Use the three numbers from the free check above. Then review them together at six months.
Often, yes. It’s worth it if customers search for your service and one new client brings in good money. Google says many small local businesses can handle the basics themselves. Start there, then hire help when you want to grow faster.
Is SEO still worth it with Google’s AI Overviews?
Yes, but the results look different now. Fewer searches lead to a click. So judge SEO by impressions, brand searches and enquiries, not just traffic. Pages that answer simple facts lose the most. Pages that help people choose or buy lose the least.
How long does SEO take to work?
Plan for 6 to 12 months to see clear results. Google can add a page to its index within a day. But earning good rankings takes much longer.
Should I cut my SEO budget if my traffic is down?
Not because of traffic alone. First check three things: Search Console impressions, searches for your brand name, and how new customers say they found you. If all three are falling, change your plan. If they’re steady or rising, SEO is likely still working.
Is Google Ads better than SEO?
Google Ads works faster and is easier to measure. But leads stop the day you stop paying. SEO is slower, but its results keep building over time. Most businesses do best with both: ads for now, SEO for next year.
What is a “zero-click” search?
It’s a search where the person gets their answer on Google and doesn’t visit any website. In early 2026, about 68% of Google searches in the US ended this way.
Can any agency guarantee a #1 ranking on Google?
No. Google clearly says nobody can guarantee a #1 ranking. If an agency promises one, that’s a warning sign.
How can I tell if my SEO is working?
Track three numbers every month: Search Console impressions, searches for your brand name, and how many customers say they found you on Google or an AI tool. If these go up together, your SEO is working.
The bottom line
SEO is worth it if people search for what you sell, your website turns visitors into customers, and you can wait six months. Just judge it by visibility and enquiries, not traffic.
SEO isn’t worth it yet if you need leads this quarter or have nothing unique to say. Anyone who tells you otherwise is just trying to sell you something.
Get a straight answer for your business
Our SEO service starts with the exact check in this guide. If the numbers say SEO isn’t right for you yet, we’ll tell you. No pressure, no fake promises.
Open a private browser window tonight and Google your own business name. Not a service, not a location. Just the name on your invoices.
If you see an AI-written summary above or between your results, you are looking at the biggest shift in brand search this year. In September 2026, Google started showing AI Overviews on most brand-name searches, which until recently were the one corner of search that stayed mostly untouched. This post covers what the data says, what it does not say, and a short check you can run yourself before spending money on anything.
What actually changed in September
A branded search is anyone typing your company or product name into Google. These are your warmest visitors: they already know you and are looking for your website, your phone number or your reviews. Until now, Google mostly answered these with the usual results and a knowledge panel.
That changed fast. Three separate datasets, all published within days of each other, point the same way:
Source
What it measured
Finding
DemandSphere (daily tracking, all markets and devices)
Share of tracked branded queries showing an AI Overview, Sept 1–29
26.12% on Sept 1, peaking at 90.48% on Sept 27, 82.06% on Sept 29
Ahrefs (232.8M US desktop crawls, Jul–Sep)
Branded vs non-branded queries with an AI Overview
Branded rose from 61.3% to 65.1% to 73.3% month by month; 83% on a late-September panel of 100 brands
Chris Long, Nectiv (manual test)
100 large enterprise brands
AI Overviews appeared for 93 of the 100
The numbers differ because the methods differ: a daily tracker, a huge crawl, and a hand-checked list of famous brands. Treat the exact percentage loosely. The direction is the story, and it is the same in all three. Search Engine Land covered the DemandSphere surge, and Ahrefs published its own crawl study within days.
What the data does not tell you
Here is the part most headlines skip. None of these studies measured clicks. DemandSphere counts an AI Overview as present whether or not your brand is mentioned in it, and the Search Engine Land report states plainly that the effect on branded click-through is unknown. Google had made no announcement about the change at the time of writing.
So nobody can honestly tell you yet that your brand traffic will fall by X%. Anyone who quotes a number is guessing. What we do know is narrower and still useful:
Google is now writing a short answer about your business on a large share of searches for your name.
That answer is assembled from sources you may not control.
Placement is unsettled. Ahrefs reports most branded AI Overviews in the first position, while Chris Long's test saw many sitting mid-page below the official site.
Why brand searches deserve more attention than they get
Most SEO reports focus on non-branded keywords, the generic terms where you compete with strangers. That makes sense for growth. But branded searches are where the sale is often closed. Someone hears about you from a friend, a podcast or a proposal, then searches your name to check you are real. What they see in those few seconds decides whether they contact you or move on.
That is why this change matters even though the click impact is unmeasured. Nobody is losing a ranking. What is shifting is who gets to describe your business first: you, or a machine summarising whatever it found.
Where the AI answer comes from
Ahrefs looked at which external sites these overviews cite. Wikipedia appeared in 71.9% of the external references, YouTube in 38.6% and LinkedIn in 22.8%. Only 3.4% of the overviews explicitly mentioned a competitor.
Read that as a business owner. The main risk is probably not that Google sends your customers to a rival. It is that a summary of you gets built from a LinkedIn page you abandoned in 2021, a directory listing with the wrong phone number, or an old press article. When a prospect sees that summary before your website, it becomes your first impression.
The 10-minute check you can run yourself
You do not need a monitoring tool for this. You need a browser and a notepad.
Search your exact business name in a private window. Note whether an AI Overview appears and where.
Repeat with your name plus "reviews", your name plus "pricing", and your name plus "contact". These are the three things customers actually look up.
Read every sentence of any overview. Mark anything wrong: services, location, opening hours, pricing, founders, history.
Click each cited source. Write down which sites feed the answer.
Fix the sources, not the summary. Update the listing, profile or page that holds the wrong fact. You cannot edit Google's answer directly, but you can correct what it reads.
Repeat monthly. Overviews change, and the September jump shows they can change within days.
In our experience with small-business sites, the errors are almost never exotic. They are stale profiles, inconsistent business descriptions across directories, and a website that never states plainly what the company does, who it serves and where. Those are fixable in an afternoon.
How to see whether your own traffic is moving
Google Search Console now has a branded queries filter in the Performance report, which splits searches for your name from everything else. Open it, compare branded clicks and impressions for August against October, and keep the figures somewhere safe.
Two cautions. Google counts AI Overview appearances in the normal Web search data and does not break them out, so you cannot see directly whether an overview caused a change. And a dip in branded clicks is not automatically bad news if enquiries and sales hold steady. Judge the trend by leads, not by the click graph. If your reports still lead with traffic, our post on which SEO metrics to measure when AI takes the clicks explains what to ask for instead.
Three questions to put to your SEO provider
If you pay an agency or freelancer, this week is a fair time to ask:
Have you checked what Google's AI Overview says about our brand, and is it accurate?
What do our branded clicks and impressions look like in Search Console since the start of September?
Which of our reported numbers measure leads, and which only measure traffic?
A good provider can answer all three in a day. If the answer is a pitch for a new tool, that tells you something too.
What to do, and what not to buy
Worth doing now (cheap):
Make your own site the clearest source about you. A plain About page and service pages that state what you do, for whom and where. This is core SEO work, not a new discipline.
Align your profiles. Same name, description, address and phone across Google Business Profile, LinkedIn, directories and social accounts.
Publish proof that is yours. Original case studies, pricing explainers and FAQs written by people who do the work. Our content creation work starts from exactly this.
Keep the site technically clean. Fast, crawlable pages with accurate structured data give any system less room to misread you. That sits under website development.
Not worth buying yet:
A monthly "AI brand monitoring" subscription for a single-location business. The free check above covers it.
Anyone promising to "control" or "rank you in" AI Overviews. Nobody can guarantee what Google writes.
Panic ad spend on your own brand name. Test first. The clicks may not have moved.
The honest summary: this is a monitoring-and-hygiene problem today, not a crisis that needs a new budget line. That could change if click data shows real losses, and we will update this advice when it does.
Frequently asked questions
Do AI Overviews really appear on brand-name searches now?
On most tracked ones, yes. DemandSphere measured them on more than 80% of tracked branded queries by late September, up from about 26% on September 1. Rates vary by brand and market, so check your own name.
Will this reduce visits to my website?
Unknown. The studies measured how often overviews appear, not how many people click. Watch your Search Console branded filter and your enquiry numbers over the next few weeks.
Can I edit or remove the AI Overview about my business?
Not directly. You can correct the pages and profiles it draws from, and wrong information often changes once those sources do. There is no setting that guarantees the wording.
Which sources do these overviews cite most?
In Ahrefs' study of external citations, Wikipedia led at 71.9%, then YouTube at 38.6% and LinkedIn at 22.8%. That is why an out-of-date LinkedIn page matters more than most owners expect.
Do I need a paid tool to track this?
Not for a small business. Searching your name, plus "reviews", "pricing" and "contact", once a month and reading the sources covers most of the value. Paid tracking makes sense for multi-brand or high-volume companies.
The takeaway
Google now writes an answer about your business on most searches for your name, and the click impact is not yet measured. Check what it says, fix the sources behind any errors, and track branded performance in Search Console. Do not buy a fix for a problem nobody has quantified.
If you would like a second pair of eyes on what Google says about your business, get in touch and we will run the check with you.
Your monthly SEO report lands. Organic traffic is down 18% on last year. Your agency says rankings are “stable” and that it’s an industry-wide AI thing. Meanwhile the phone is ringing about as often as it used to — or maybe it isn’t, and nobody has checked.
That confusion is the real story in search right now. Google’s AI answers are absorbing clicks that used to show up in your traffic chart, which means the SEO KPIs most businesses were sold — traffic, rankings, domain authority — have quietly stopped telling you whether SEO is working. This post is for the person paying the invoice: what changed, which numbers still mean something, and which ones you can stop paying anyone to report.
Why your traffic chart stopped telling the truth
Four independent sets of data published in the last few weeks point the same way:
AI Overviews are now the norm, not the exception. Comscore’s Q2 2026 data shows an AI Overview on 39.4% of US desktop Google searches in June, up from 25.8% a year earlier. Yet total US desktop searches rose 8% over two years, to 77 billion a quarter. People are searching more and clicking less.
AI Mode cuts them further. A preregistered field experiment with 1,100 US Chrome users (Wang, Gleason, Bart, Wilson and Metaxa, August 2026) found that pushing searches into Google’s AI Mode cut the share of searches leading to an outside website by 18.8 percentage points.
The effect shows up in real sites’ data. When AI Overviews launched in France on 22 July 2026, Ahrefs tracked 963 domains in Search Console: sites with heavy AI Overview exposure saw click-through rates fall 23.1% within weeks.
Google disputes the framing. Its Head of Search, Liz Reid, has said total organic click volume is “relatively stable” year on year and that the clicks Google sends are higher quality — though Google hasn’t published the numbers behind that. Both things can be true: fewer people click, and the ones who do are further along in deciding. For a business, that’s not necessarily bad news. It just means traffic is now a poor proxy for results, and a report built around it will mislead you in both directions.
The SEO KPIs that still mean something
Most “SEO KPI” lists give you 11 or 16 metrics. That’s useful if you do SEO for a living. If you’re the one paying for it, you need about five, and only one of them pays the bills.
KPI
What it tells you
Where it lives
Watch out for
Leads or sales from organic search
Whether SEO is producing revenue
GA4 key events by channel, plus your CRM
Untracked phone calls and form tools that don’t fire events
Organic conversion rate
Whether the fewer, later-stage clicks are turning into enquiries
GA4 (key events ÷ organic sessions)
A rising rate on falling traffic is often a good sign, not a bad one
Branded search demand
Whether more people know your name and look you up
Search Console, queries containing your brand
Seasonality — compare year on year, not month on month
Non-branded impressions on money pages
Whether Google still shows your service pages to new buyers
Search Console, Pages tab
Blog impressions inflating the total
Lead quality
Whether organic enquiries are the kind you want
Your CRM or a “how did you find us?” field
Nobody asking the question
Notice what’s missing: total sessions, keyword rankings and click-through rate. They’re still useful to whoever does the work, as diagnostics. They just shouldn’t be the headline of a report sent to someone judging whether the spend is worth it.
The metrics you can stop paying for
Some of what shows up in monthly SEO reports was always padding. AI search has made it worse than padding, because it now actively misleads.
Rank tracking for hundreds of keywords. You can rank first and still get few clicks if an AI Overview sits above you. A ranking without the business outcome next to it is a vanity number.
Domain Authority or Domain Rating. These are third-party tool scores, not Google metrics. Useful for comparing sites; meaningless as proof that your money is working.
“Total traffic is up.” Traffic from a blog post answering a trivia question in another country doesn’t pay rent. Ask which pages and which visitors.
A paid “AI visibility score” as a headline KPI. For most small and mid-sized businesses this isn’t worth a separate subscription yet — and the data behind most of them is still thin. Google’s own new AI report in Search Console is free.
If your current report is mostly the four items above, the honest answer is that you’re paying for a report, not for insight. A good SEO service should be able to tell you in one sentence what organic search produced for your business last quarter.
A 10-minute check you can run yourself
This is the first thing we look at when a business owner asks whether their falling traffic is a problem. You need access to Google Search Console and Google Analytics 4, both free.
In Search Console → Performance → Search results, set the date range to the last 16 months and tick both Total clicks and Total impressions. Google’s documentation confirms AI Overviews and AI Mode are counted inside this “Web” report, so this is your whole picture.
Add a query filter that does not contain your brand name. Note whether non-branded impressions are flat, rising or falling, and whether clicks are moving with them.
Switch the filter to queries that do contain your brand name. That’s your branded demand line.
In GA4 → Reports → Acquisition → Traffic acquisition, find the Organic Search row and read the Key events column for the same periods. If that column is empty, your enquiries aren’t being tracked — fix that before judging anything else.
Compare the result with the table below.
Impressions
Clicks
Organic leads
What it probably means
Flat or up
Down
Steady or up
AI answers are absorbing early-stage clicks. SEO is still working. Stop worrying about the traffic chart.
Flat or up
Down
Down
Visibility is fine but pages aren’t converting the visitors who do arrive. That’s often a website problem, not an SEO problem.
Down
Down
Down
A genuine visibility loss: a competitor, an algorithm update or a technical issue. This needs a proper diagnosis.
Up
Up
Flat
You’re attracting the wrong audience, usually through informational blog content. Rethink topics, not volume.
The first row is the most common pattern right now, and it’s the one most likely to trigger a panicked, expensive reaction. If that’s you, the right move is usually to do less of what inflates traffic, and more of what converts the traffic you have.
Five questions to ask your SEO agency this month
How many leads or sales came from organic search last quarter, and how does that compare with last year? If they can’t answer, tracking is the first job.
Which of our pages produce those leads? The answer should be a short list of service and product pages, not the blog.
Are our non-branded impressions growing on the pages that sell? This is how you separate “AI took the clicks” from “we lost visibility”.
What are you changing because of AI Overviews, and why? A good answer is specific. “We’re optimising for AI” isn’t.
What would you stop doing if the budget were cut by a third? This tells you which activities they believe actually move the numbers.
What this means for where the money goes
If AI answers now handle the “what is” and “how does” questions, the value of publishing lots of generic explainer posts has dropped. They attract traffic that increasingly never arrives, and when it does, it rarely buys.
The money is better spent where the remaining, higher-intent clicks land. That means service pages that answer the buying questions (price ranges, process, timelines, proof), fast pages that make enquiring easy, and a handful of genuinely original pieces that give AI systems and people a reason to cite you. That’s a mix of website development and focused content creation, not more of either for its own sake.
It also means accepting a smaller, more honest traffic number. A site getting 30% fewer visits and the same number of enquiries is converting better than it was, not failing.
Frequently asked questions
What are the most important SEO KPIs for a small business?
Leads or sales from organic search come first, followed by organic conversion rate, branded search demand, non-branded impressions on your service pages, and lead quality. Rankings and total traffic are useful diagnostics but poor headline KPIs now that AI answers absorb many clicks.
Why is my organic traffic down if my rankings haven’t changed?
Most likely because an AI Overview or AI Mode answer now sits above your result and answers the question without a click. Pew Research found clicks on traditional results fall from 15% to 8% of visits when an AI summary appears. Check whether your impressions and leads held steady before assuming something is broken.
Does Google Search Console show traffic from AI Overviews?
Yes. Google says AI Overviews and AI Mode appearances are included in the Performance report under the Web search type. Its separate generative AI report, available to all sites since August 2026, shows impressions for AI features but not clicks.
Should I pay for an AI visibility tracking tool?
For most small and mid-sized businesses, not yet. Start with the free Search Console data and with tracking enquiries properly. A paid tracker makes sense once you have a clear question it answers that your existing data can’t.
How often should I review SEO KPIs?
Quarterly is enough for decisions, compared year on year to remove seasonality. Monthly reports are fine for spotting technical problems early, but judging SEO on a single month’s traffic leads to overreaction.
The takeaway
AI search hasn’t made SEO pointless. It has made traffic a bad scoreboard. Judge your SEO on the enquiries and sales it produces, use impressions and branded demand to tell a visibility problem from an AI-clicks problem, and stop paying for reports that lead with rankings and domain scores. If a report can’t tell you what organic search earned you last quarter, it isn’t a report worth paying for.
Want a second opinion on your current SEO reporting? Get in touch with Digital Roof and we’ll run the check above on your own data and tell you plainly what it shows, including if the answer is that your current setup is fine.
For the first time in the internet's history, automated traffic overtook human traffic. Cloudflare's own traffic data confirmed it in June 2026: bots now account for the majority of web requests worldwide, a milestone that arrived roughly 18 months earlier than the company itself had predicted. Every SEO vendor immediately built a webinar around that number.
But the figure that actually matters for a small or mid-size business is smaller, and far more specific. AI agents built to browse, compare, and act on behalf of a real person made up an estimated 4.2% of HTML requests on the average website in 2025, according to Cloudflare's own traffic data. Most of that 57% "bots beat humans" headline is old-fashioned scrapers, uptime monitors, and search crawlers that have been part of the web for years. The AI-agent slice is much smaller today, but it's growing faster than anything else on the internet.
That's the real question behind "agentic SEO": not the scary headline, but what your website needs to do when the visitor filling out your contact form, or comparing you to a competitor down the road, isn't a person at all.
What "Agentic SEO" Actually Means
It helps to separate two things that get lumped together under "AI and SEO."
AI visibility, or generative engine optimization (GEO), is about whether ChatGPT, Gemini, or Perplexity mention your business when someone asks a question in a chat window. Digital Roof has covered that ground before — it's about being cited.
Agentic SEO is a different, newer problem. It's about what happens after a person hands a task to an AI agent — Claude, ChatGPT's browsing mode, Perplexity's Comet browser, Gemini's agent mode — and that agent goes and actually visits your website to complete it. Booking an appointment. Requesting a quote. Comparing your services against three competitors' sites in the space of a few seconds. The agent isn't just reading your homepage the way a search crawler does; it's trying to click, fill in, and submit things on your behalf.
If your site was only ever built and tested with a human clicking through it, that's the part worth checking.
The Agents Are Already Here — Here's How to Check
You don't need server access to see this for yourself. Open a private browser window, go to ChatGPT, Claude, or Perplexity, and ask a simple question: "What does [your business name] do, and what's the best way to book with them?" If the answer is wrong, badly out of date, or the assistant can't find you at all, that's not just a visibility problem — it's a sign that an agent landing on your site right now would struggle to do anything useful once it got there.
The scale of this shift is well documented. HUMAN Security's 2026 State of AI Traffic report, based on more than a quadrillion tracked interactions, found agentic AI traffic grew by roughly 7,851% year over year — automated activity expanding about eight times faster than human traffic. Separately, network analytics firm 51Degrees tracked more than 3 billion website visits through May 2026 and found AI "harvester" bots — agents identifying themselves as ClaudeBot, PerplexityBot, and similar — rose from 8% to 13% of all web sessions in a single year.
What's growing
Figure
Source
All automated (bot) traffic vs. human traffic
~57% bot / ~43% human, worldwide
Cloudflare Radar, June 2026
AI agent traffic specifically (share of HTML requests)
~4.2% average across 2025
Cloudflare Radar
AI "harvester" bot sessions
8% → 13% in one year
51Degrees, 3B+ visits analyzed
Year-over-year growth in agentic AI traffic
~7,851%
HUMAN Security, 2026 State of AI Traffic report
None of this means your local service business is about to be overrun by robot customers next month. It means the trend line only points one way, and the businesses that fix the basics now won't be scrambling later.
What Your Website Actually Needs Right Now
The useful part of "agentic SEO" doesn't require a developer or a new budget line. It's five habits — adapted from Moz's own agent-readiness checklist for a business audience — and most of them are things a decent SEO services engagement should already be catching.
Give agents something to actually do. A page that only explains your services in a paragraph gives an agent nothing to act on. A booking form, a quote request, a "check availability" button — anything a visitor (human or agent) can complete in one or two steps — is what turns a visit into a result.
Make your information consistent everywhere. If your phone number, hours, or service area differ between your homepage, your Google Business Profile, and a directory listing, an agent has no way to know which one is correct — so it often just gives up or guesses wrong. Consistency across every listed source is one of the simplest wins in SEO work, and it matters even more here.
Write down anything you'd normally explain out loud. If a customer would normally have to call and ask "do you serve my suburb," "what do I need to bring," or "how does the process work," that answer belongs in plain writing on the site. This is squarely content creation work, not a technical fix.
Use real button labels, not "click here." "Book a consultation" or "request a quote" tells both a screen reader and an AI agent exactly what will happen next. Vague link text is one of the oldest accessibility complaints in web design, and it's now also an agent-readability problem.
Keep forms simple enough to complete without a phone call. A contact form that requires ten fields, a CAPTCHA with no fallback, or a multi-step wizard is hard enough for a busy human — it can stop an agent completely. If your site was built years ago and never revisited, this is worth a proper look through website development.
What You Can Safely Skip For Now
Some of what gets marketed as "agentic SEO" is genuinely premature for most businesses, and it's worth saying so plainly.
Google announced WebMCP — a new browser standard letting websites expose forms and functions directly to an AI agent inside Chrome — at its I/O developer conference in 2026. It's currently running as an experimental origin trial starting in Chrome 149, with companies like Expedia, Shopify, Etsy, and Target testing early implementations. The underlying specification is still being worked out through a W3C working group, and full standardization is expected to take years, not months.
The same goes for full compliance with the Agentic Commerce Protocol that OpenAI, Stripe, and the major payment networks are building for AI-driven checkout, or standing up your own MCP server so agents can query your business data directly. Those are real developments, but they're solving problems at the scale of national retailers and payment platforms. A single-location service business chasing full protocol compliance before fixing an inconsistent phone number or a ten-field contact form is solving next year's problem before this year's.
If an agency pitches an "agentic SEO audit" that's heavy on protocol names and light on your actual booking flow, that's a reasonable thing to push back on.
A Simple Starting Point
Run the AI-answer check above on your own business today, and note anything that's wrong or missing.
Fix inconsistent information wherever you find it — website, Google Business Profile, and directories.
Test your main conversion action (booking, contact, or quote request) the way an agent would: can it be completed with just a name and email, in one pass, without a phone call or a CAPTCHA dead end?
None of that requires new software. It requires treating the next few months the way good SEO practice has always treated any traffic shift — with a plain audit, not a panic purchase.
Frequently Asked Questions
What is agentic SEO? Agentic SEO is making sure AI agents acting on a person's behalf — booking, comparing, filling out forms — can actually complete tasks on your website, rather than just reading it the way a search engine crawler does.
Are AI agents actually visiting my website right now? Some are. Cloudflare's own traffic data put AI agent activity at roughly 4.2% of HTML requests across 2025, and independent tracking from 51Degrees and HUMAN Security both show that share growing quickly. It's a minority of your traffic today, but the fastest-growing part of it.
Do I need to worry about WebMCP or the Agentic Commerce Protocol? Not yet, for most small and mid-size businesses. Both are still in early, opt-in stages, tested mainly by large retailers and platforms. They're worth knowing about, not worth budgeting for before the basics — accurate information, working forms, clear labels — are handled.
How is agentic SEO different from AI visibility (GEO)? AI visibility is about whether a chatbot mentions your business in its answer. Agentic SEO is about what happens next — whether an agent that then visits your actual website can complete a task there. One is about being found; the other is about being usable once found.
What's the one thing I should do first? Ask an AI assistant what it knows about your business and whether it can tell someone how to book with you. Whatever it gets wrong is your starting checklist.
Agentic SEO isn't a rebrand of AI visibility, and it isn't a reason to buy into every new protocol the moment it's announced. It's a reminder that some share of your website's visitors — right now, growing every month — aren't people, and a site that only works for a patient human with a mouse is quietly turning some of them away. Fix the five basics above, skip the enterprise-scale protocols for now, and revisit the question in six months rather than six years.
If your site's booking flow, contact forms, or on-page information could use a proper audit before that traffic shift accelerates further, get in touch with Digital Roof and we'll walk through what actually needs doing.
If you run a business, you've probably heard both sides of this argument in the same week. One agency tells you AI content will tank your rankings. Another tells you to publish three AI-written posts a day because "Google can't tell the difference." Both are selling you something, and neither is telling you what the actual 2026 data shows.
Here's the real story: content marketing results just hit a 12-year low, at the exact moment AI adoption hit an all-time high. That's not a coincidence, but it's also not the story it looks like at first glance — and understanding the difference is worth more to your business than any AI writing tool.
The Data: Content Marketing Just Hit a 12-Year Low
Every year, Orbit Media surveys more than 1,000 content marketers about what they're doing and how well it's working. Its 2026 Blogging Statistics report, based on 1,042 responses, found that only 14% of marketers now say their blog delivers "strong results." That's the lowest figure in the report's 13-year history — six points below the previous low, and nearly half of the 26% who said the same in 2022. Search Engine Land's coverage of the report called it a "wake-up call," quoting MarketingProfs chief content officer Ann Handley on marketers needing to be more deliberate about which practices they abandon.
At the same time, AI adoption among content marketers hit 92.4%, up from virtually nothing three years ago. The average blog post now takes 3 hours and 20 minutes to produce, down from over 4 hours in 2022 — a saving of roughly 50 hours a year per marketer.
So marketers are producing content faster than ever, with more help than ever, and getting worse results than they have in over a decade. If your own content hasn't been performing the way it used to, you're not imagining it, and you're not alone.
What Google Actually Says About AI Content (Not What You've Heard)
Before you conclude that AI is the villain here, it's worth checking what Google itself actually says — because it's not what most of the fear-based sales pitches claim.
Google's own Search Central documentation on generative AI content states plainly that using automation, including AI, to produce content is not against its guidelines on its own. What violates Google's spam policies is using automation — AI or otherwise — "with the primary purpose of manipulating ranking in search results," or publishing content at scale without adding value for users. Google has used automation to generate weather reports and sports scores for years without penalty. The method was never the issue. The value to the reader is.
That single distinction matters more than most content advice you'll read this year, because it means the honest answer to "is AI content bad for SEO?" is no — not by itself. Which means something else is driving the 12-year-low results, and the Orbit Media data tells us exactly what it is.
So Why Are Results Falling If AI Isn't the Problem?
The survey didn't just measure the decline — it measured which specific practices correlate with strong results, and then tracked how often marketers are still doing them. The pattern is consistent: the practices most linked to success are the ones marketers have quietly abandoned.
Practice
Impact on results
What's changed
Collaborating with outside experts
2.6x more likely to report strong results
Fell from 25% of marketers (2017) to just 7% (2026)
Original research
Roughly 50% improvement in reported results
Declining after a decade of steady growth
Formal human editing
Nearly 2x more likely to report strong results
Increasingly replaced by AI-only review
Keyword research
Positive correlation with results
Use is converging toward "sometimes" instead of "always"
Paid content promotion
Positive correlation with results
Declining as a standard practice
Consistent analytics review
Positive correlation with results
Data-driven marketers report more success than those who "eyeball it"
None of these six things are about whether AI touched the content. They're about whether a human being added expertise, checked the facts, edited the draft, and tracked what actually happened afterward. AI adoption removed the time cost of writing — and, for a lot of businesses, it removed the discipline that used to come attached to that time cost. The content got faster to produce. It didn't get more useful to the person reading it, and Google's own documentation is explicit that usefulness to the reader is the only thing it's actually grading.
The 10-Minute Audit You Can Run on Your Own Blog
You don't need a tool subscription to find out where your own content sits on this. Pull up your last 10 published posts and check each one against three questions:
Did a human read and edit this before it published — not just skim it, but actually fact-check it and cut what didn't need to be there?
Does it contain anything an AI model couldn't have generated on its own — a client example, an original data point, a named expert's opinion, a result you measured yourself?
Do you know how any of these 10 posts actually performed — traffic, rankings, leads — or did you publish and move on?
If you answered "no" to two or more of those questions across most of your last 10 posts, you've just found your own version of the Orbit Media data. It's not an AI problem. It's a process that skipped the parts that were always doing the actual work.
What to Do Instead
The fix isn't to stop using AI, and it isn't to publish less. It's to put the abandoned steps back into whatever process you're already using:
Keep a real human edit as a required step, not an optional one — someone who checks facts, removes filler, and can explain why each paragraph is there.
Add one thing AI cannot supply: a client story, an original number from your own business, a quote from someone with direct experience. This is also what separates content that gets cited by AI search tools from content that doesn't, since those tools are themselves looking for a source with something to add.
Don't cut keyword research or promotion just because writing got faster. Those steps were never about the writing — they're about whether the right person ever sees what you wrote.
Track what you publish. Even a simple monthly check of which posts brought in traffic or inquiries tells you more than any AI detector ever will.
When "Don't Buy This" Applies
Be skeptical of two opposite pitches right now. The first is any agency or tool promising that fully AI-generated content, published with no editing pass, will keep ranking indefinitely — the Orbit Media data says that exact combination (heavy AI use, declining human editing) is precisely where results are falling. The second is the reverse panic: anyone telling you to purge every AI-assisted sentence from your site "before Google notices." Google's own documentation doesn't support that policy, and stripping useful content because of how it was drafted fixes nothing if the underlying process still skips fact-checking and editing.
The honest work is less dramatic than either pitch: keep using AI where it helps, keep a human in the loop where it matters, and measure the result. That's not a new idea. It's the same one Orbit Media's data shows was working better six years ago, before most teams stopped doing it.
Takeaway
AI didn't break content marketing, and Google isn't quietly penalizing every AI-assisted post you've published. What's actually happening is more ordinary and more fixable: speed replaced a set of habits — editing, expertise, research, measurement — that were doing more of the work than most teams realized. If your own results have slipped, that 10-minute audit above will usually tell you which of those habits went missing first.
If you'd rather have someone run that audit and fix the gaps than do it yourself, our content creation team builds every piece around a human editing pass and original input specific to your business — and our SEO services team can tell you which of those six factors is actually costing you the most. Get in touch and we'll tell you honestly whether the fix is process or content.
FAQ
Does Google penalize content just because it was written with AI? No. Google's Search Central documentation states that using automation, including AI, to create content is not itself against its guidelines. Its spam policies target content produced primarily to manipulate rankings or published at scale without adding value — not the use of AI as a tool.
Why did content marketing results drop in 2026 if AI is supposed to make marketing more efficient? Orbit Media's 2026 survey of 1,042 marketers found that the practices most correlated with strong results — expert collaboration, original research, human editing, keyword research, paid promotion, and analytics review — all declined over the same period AI adoption rose to 92.4%. The drop in results tracks the drop in those practices, not AI use itself.
Should I stop using AI to write content for my business? No. There's no evidence that avoiding AI improves results, and Google doesn't penalize its use. The data points to a different fix: keep a human editing and fact-checking every piece, and make sure each one includes something AI can't generate on its own, like a real client example or original data.
What's the single highest-impact change if my content isn't performing? Based on the Orbit Media data, formal human editing and outside expert input have the strongest correlation with reported success — marketers who use them are roughly twice as likely to call their results "strong." If you can only fix one step in your process, start there.
How do I know if my content actually needs a human editing pass? Run the 10-minute audit above on your last 10 posts. If most were published without a dedicated edit for accuracy and relevance, or without anything unique to your business in them, that's the gap to close first — regardless of whether AI was involved in the first draft.
