Can AI Do SEO? What Business Owners Should Still Pay For
AI can now audit, sort and draft. It still can't fix your website or own your results. Here's what to stop paying for, and what's still worth it.
On this page
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.
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.
Audit exports, keyword lists, generic blog posts and reports that only restate numbers.
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.
Open any AI chatbot and ask: "Do a full SEO audit of [your website] and list the top 10 problems."
Then ask: "Which exact pages did you fetch, and which tool measured each problem?"
Open Google Search Console (free) and look at the Pages report and the Core Web Vitals report for the same site.
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:
It shows up in Search Console or on the page itself.
Advice that would apply to any website.
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.
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.
Send Us Your SEO ReportSources
- Semrush, How to do an SEO audit with Claude & the Semrush MCP, 15 October 2026
- Ahrefs, Si Quan Ong, How to Automate Content Creation: A Follow-Along Guide, 16 October 2026
- Search Engine Land, What the PPC agency of 2027 will look like, October 2026
- Król, K. (2025), Between Truth and Hallucinations, Applied Sciences 15(5), 2292
- Google Search Central, Guidance on using generative AI content
- PPC Land, Google's 1 October 2026 update to its AI content guidance
Keep Reading
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.