We Audited 134 SEO Agency Websites for AI Search

SEO agencies and freelancers keep their own websites in better shape than the average business does, but only on the checks that have been on SEO checklists for a decade. We audited 134 websites that sell SEO services, between 23 July and 19 October 2026, and compared them with 218 other sites that went through the same audit in the same period. The SEO sites scored a median of 86 out of 100 against 82. On the newer checks that decide whether an AI assistant can quote a page, they did worse than the businesses they sell to: 66% have no llms.txt file (against 51%), and 37% ask questions in their headings without answering them directly underneath (against 21%).

So the profession that's now selling "get found in ChatGPT" mostly hasn't done it at home yet. It's a useful map of where the work is, for agencies and for the businesses hiring them.

If you'd like your own site's numbers first, run the free SEO audit. It's the same check all 352 of these sites went through, it takes about 30 seconds, and there's no signup.

Who was in the study?

The SEO group is 134 websites whose own homepage title, description or main heading offers SEO services. That's freelancers, consultants and agencies, from one-person practices to firms with teams. We read every one of the 138 sites the filter caught and removed four that don't sell SEO services: two SEO software products, a virtual-assistant agency, and our own site.

By domain ending, 21 of the 134 are Indian (.in), 11 British, 11 Australian and 5 German, and the remaining 86 are mostly .com.

The comparison group is the 218 other websites audited over the same stretch: local service businesses, online shops, SaaS products, clinics, consultants. Same instrument, same checks, same scoring version. That comparison is what earlier studies of agency websites didn't have, and it changes the story. Without it you can only say "agencies have problems." With it you can say which problems are specific to agencies.

Two limits you should know before reading any further. These sites weren't chosen at random: many came from our own outreach lists, because SEO freelancers and agencies are who we sell to. And the audit reads the homepage only, so everything below describes homepages.

Where do SEO agency websites beat everyone else?

On the classic on-page basics, by a clear margin. The meta-tags category averaged 71.9 out of 100 for SEO sites against 64.8 for everyone else, and headings averaged 72.7 against 67.6.

These are the checks every audit tool in the industry has flagged for years, and the checks agencies run on client sites every week. It shows. The SEO sites are noticeably better at the work they've been doing longest.

Where do they fall behind?

On AI readiness and on page weight. Here the gap runs the other way.

The question-heading result is the one we'd act on first. Over a third of SEO homepages ask a question in a heading ("Why choose us?", "What does SEO cost?") and then don't answer it in the first sentence or two below. AI assistants and Google's AI Overviews tend to quote the short passage that settles a question. A heading followed by a paragraph of build-up gives them nothing tidy to lift. The check arrived on 23 July, so we measured it only on the 132 SEO sites and 216 others audited after that date.

The llms.txt gap is real but needs context. It's an optional plain-text file that tells AI assistants what a business does and which pages matter. Not every AI engine reads it yet, so a missing file is a missed easy win and not a broken site. What stands out is the direction. The people selling AI-search services are less likely to have one than a typical plumber's or online shop's website in the same sample. We explain the file and whether you need one in what llms.txt is, in plain English, and you can make one in a minute with the free llms.txt generator.

The speed numbers have a likely explanation that we didn't measure. Agency sites tend to carry chat widgets, tracking pixels, heatmap tools, booking embeds and page-builder plugins, and each one adds a script. That's our guess at the cause. The finding itself is just the count: 63% of SEO homepages load at least one script in the page header that makes the browser wait before showing anything, against 53% of other sites.

Some things are just as broken everywhere

Heading order is the clearest. 51% of SEO sites have headings that skip levels, jumping from a main heading straight to a sub-sub-heading, and so do exactly 51% of the comparison group. One in five SEO homepages (20%) has no main heading at all, against 23% elsewhere, and 13% have more than one.

That matches what we found in our earlier study of 226 small-business websites, where heading structure was the worst-scoring category overall. Visual size carries meaning for people, so a skipped level looks fine on screen. To a machine trying to work out which passage answers which question, the outline is broken. Site builders make this easy to get wrong, because people pick heading styles by how they look.

What about agencies that sell AI search specifically?

Their sites are in better shape, but the group is small. 24 of the 134 mention AI search, GEO, AEO, ChatGPT or similar on their homepage. Of those, 42% (10 of 24) have no llms.txt file, which is much better than 66% for SEO sites overall. Only 29% have headings that skip levels.

With 24 sites, one or two either way moves those percentages by several points, so treat this as a direction and not a measurement. Speed didn't improve: 63% of this group also had scripts delaying the page.

The one earlier study of this question that we could find, an audit of 12 UK SEO and AI-search agency homepages run on 26 April 2026, found 4 of the 12 publishing llms.txt and concluded that most agencies sell AI search optimisation without implementing it on their own sites. Our sample is about eleven times larger, spans several countries and includes a comparison group. It agrees on llms.txt, and it adds the part a 12-site study can't show: on the older checks, agencies are clearly ahead.

What should an agency fix on its own site first?

Fix the question headings, then the scripts, then publish llms.txt. That's ordered by how often each problem appeared and how little it costs to fix.

Answer every question heading in its first sentence. Go through your homepage and service pages. Under every heading phrased as a question, make the first one to three sentences the actual answer, then expand. This takes an afternoon and costs nothing.

Defer the scripts in your page header. Add defer or async to scripts that don't need to run before the page shows, or move them to the end of the page. Most WordPress caching and performance plugins have a setting for this. While you're there, remove any widget nobody uses.

Publish an llms.txt file. Write a short plain-text summary of what you do and list your key pages. It's ten minutes of work, and it's the most visible signal to any prospect who checks whether you do this yourself.

Then check your heading order and give every page exactly one main heading. Our AI crawler checker will also confirm that ChatGPT, Perplexity and Google's AI crawlers can reach your site. 4% of SEO homepages in this study block AI assistants outright.

There's a commercial reason to do this before your next pitch. Prospects increasingly run a free audit on the agency they're about to hire. If you plan to sell AI-visibility work, your own site is the first case study they'll look at. We wrote up how to package that service honestly in how an agency can sell AI-visibility audits.

Methodology, in enough detail to check us

Instrument. Every figure comes from the public RankAgent audit, running the same checks with the same scoring version (v2, live since 21 July 2026) on every site. Nothing was hand-scored or re-run to improve a number.

Window and unit. The SEO group's audits ran from 23 July to 19 October 2026. The comparison group covers v2 audits from 21 July to 28 October 2026. Where a site was audited more than once, we kept its most recent completed audit, after normalizing the host (dropping a leading www). The unit is one homepage per site.

Classification. A site is in the SEO group if its live homepage title, meta description or main heading contains "SEO" or "search engine optimisation", which we fetched on 29 October 2026. We then read all 138 matches by hand and removed four that don't sell SEO services. The AI-search subgroup is the SEO sites whose same text mentions AI search, GEO, AEO, ChatGPT, LLMs, answer engines, AI Overviews, AI visibility or Perplexity. The comparison group is every other v2 audit, minus placeholder domains.

Sampling. Non-random. A large share of both groups came from our own outreach lists, and the rest are public runs of the free audit. There's no weighting. These figures describe these sites, not all agencies or the whole web.

Denominators. The question-heading check started on 23 July 2026, so it's measured on the 132 SEO and 216 comparison sites audited after that. Every other figure uses the full 134 and 218.

What we didn't measure. Rankings, traffic, AI citations or client results. A site with a skipped heading level may still rank well and win plenty of business. These are statements about markup, not about how good any agency is at its job. We also haven't named any site, and we won't.

Conflict of interest. RankAgent sells SEO software to agencies and freelancers, and the instrument is our own product. Weigh the numbers with that in mind, and run your own site through the audit to check us.

Frequently asked questions

Do SEO agencies do SEO on their own websites?

Mostly, yes, for the classic checks. In our audit of 134 SEO agency websites, the median score was 86 out of 100 against 82 for other businesses, with fewer missing descriptions, schema gaps and canonical problems. The gap is in AI readiness, where agency sites trailed the comparison group.

What do SEO agency websites most often get wrong?

The most common problems were a missing llms.txt file (66%), scripts delaying the page (63%), headings that skip levels (51%), images with no description (40%) and question headings without a direct answer (37%). The last one is the cheapest to fix and the one that matters most for being quoted by AI assistants.

Is an SEO agency website audit worth doing before hiring an agency?

It's a useful five-minute check. A free audit shows whether an agency applies its own advice. Don't treat the score as a verdict on the agency's skill, since a small practice's homepage says little about its client work. Use it as a starting point for questions.

Does llms.txt matter for an agency website?

It's optional and not every AI engine reads it yet, so it won't move rankings on its own. For an agency selling AI-search services, though, it's ten minutes of work, and it's the first thing a prospect who knows the term will look for.

How were the 134 sites chosen?

They're every site in our audit records from 23 July to 19 October 2026 whose homepage offers SEO services, checked by hand. Many came from our own outreach, so the sample isn't random. The methodology section above has the full detail.

Can I see how my own site compares?

Yes. Run the free SEO audit on your homepage and you'll get the same score and the same checks used in this study, in plain English. If you run an agency, see how RankAgent works for agencies, including a white-label sample client report.