Bot access
Readable HTML
Answer-first structure
Honest measurement
Bot access
Readable HTML
Answer-first structure
Honest measurement
Two of the things you are being sold have actually been measured. Schema markup made no meaningful difference to AI citations in a study of nearly two thousand pages, and 97% of llms.txt files were never requested at all. What is left is real, narrow and worth doing — so that is what we sell.
The volume is falling and it is going to keep falling. But the visitors who still arrive from an AI answer are worth more than the ones you lost, which is a different problem from the one most agencies are selling you a fix for.
25%
Of Google searches triggered an AI Overview in Q1 2026, from a study of 21.9 million searches
15→8%
Click-through on traditional results when an AI summary appears, versus when it does not (Pew)
3×
AI-referred visitors convert, compared with other channels
4.4×
The value of an AI-referred visitor against average organic traffic
The split matters more than the average. Simple factual queries — definitions, dates, quick how-tos — take the heaviest losses, because an AI answer fully replaces the click. Comparison and decision queries hold or grow. If your content is mostly the first kind, no amount of technical work will save it, and we would rather tell you that than audit it.
Both are on almost every AEO service page you will read this week. Both have now been measured. We are naming them because you are about to be quoted for them, possibly by us if we had not checked.
Does not work
Adding JSON-LD is sold as the way to get cited by AI. It is not. The largest study on this measured it directly and found the effect on AI citations to be negligible — and found that AI systems ignore hidden schema during retrieval anyway, extracting only the visible HTML.
The evidence
Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 against 4,000 control pages, using a matched difference-in-differences analysis across Google AI Overviews, Google AI Mode and ChatGPT.
What to do instead
Schema still earns rich results in classic search, and it is cheap. Keep it, and we will fix it if it is broken. Just do not pay anyone a monthly fee for it as an AI service.
Mostly ignored
The proposed standard for telling AI models what your site is about. Publishers adopted it enthusiastically and the models did not. Of the major AI companies, only Perplexity says it uses the file at all; Google, OpenAI and Anthropic do not. Today its main real consumers are coding agents.
The evidence
Server-log analysis across 137,000 domains found that 97% of llms.txt files received zero requests in May 2026 — in the same year adoption grew 8.8×, from 4,088 files to 36,120.
What to do instead
It takes about twenty minutes to add and it may matter later. Add it if you want one. It is not a deliverable and it should not be a line on an invoice.
Four groups of four. This is the whole list, published rather than gated, because a checklist you can read before you pay is the only kind worth publishing.
01
02
03
04
This is the section nobody publishes, because it undercuts the reporting. One thing works and three things do not, and you should know which before anyone sends you a dashboard.
GA4 added a native AI Assistant channel on 13 May 2026. It recognises ChatGPT, Gemini, Deepseek, Copilot and Grok with no configuration, and it only counts forward — nothing before that date is classified.
Perplexity is not in it. Arguably the highest-intent AI source lands in Referral, and needs a custom channel group to separate out.
Google’s own AI Overviews and AI Mode are not in it either. Those clicks are counted as Organic Search — which is very likely your largest AI source, reported as though nothing changed.
An estimated 60 to 70% of AI sessions arrive with no referrer at all and disappear into Direct. No configuration recovers those.
Which is why we do not sell monitoring. A monthly retainer reporting on AI visibility would be charging you for a number that is mostly invisible. We will set the measurement up properly as part of the audit, show you exactly where the holes are, and then leave you to read it yourself.
If a competing quote promises any of these, ask how it is measured. The answer will tell you most of what you need to know about the quote.
There is no ranking to buy. Nobody controls which sources a model names, the selection is not published, and anyone selling you a position in an AI answer is selling something they cannot deliver.
The click-through decline is structural, not a penalty you can recover from. This work is about being worth citing when a citation happens — it is not a route back to 2023.
Most of the AI traffic arriving at your site is invisible to analytics. We will report what is measurable, and say plainly which part is not, rather than filling the gap with a confident-looking chart.
Largely, yes, and anyone telling you otherwise is selling something. The Ahrefs finding was that AI systems favour sites that already do the fundamentals well — good content, authority, links, real rankings. The genuinely new work is narrow: retrieval-bot access, content being in the HTML, answer-first structure, and measurement. That narrow list is what this page sells. The rest is the SEO you should be doing anyway.
It depends entirely which ones. Blocking the training bots while allowing the answering bots is now a common position, and most robots.txt files we look at have accidentally done the opposite. Whether to let models train on your content is a business decision about your content, not a technical one, and we will set it up either way — but you should make the decision deliberately rather than inherit it from a template.
We do not know, and neither does anybody else. Retrieval indexes refresh on their own schedules and none of them publish it. What we can tell you is when the technical blockers are gone, because that is verifiable in your server logs, usually within the same week.
No. See the three things above that we will not promise. If a quote you are comparing this against does guarantee them, ask how the guarantee is measured — the answer will tell you most of what you need to know.
Yes, with some overlap. The technical audit is about being crawlable, fast and indexable for Google. This covers four things that audit does not: the training-versus-retrieval bot distinction, JavaScript-rendered content, answer-first page structure, and AI measurement setup. If you have already had the technical audit, this one is smaller and cheaper.
A little, in one specific way: Webflow renders pages server-side, so your content is in the HTML rather than assembled by JavaScript after load. That is one whole category of problem you do not have, and it is the most common thing we find on React and Next.js sites. Everything else on this page is the same work on any platform.
We will tell you which AI bots can currently reach you, whether your content is readable when they do, and whether any of this is worth paying for on your site.