The five signals
Every AI engine works through the same chain, in this order. If it breaks early, nothing later in the chain can save you.
- 1. Can AI find you? Crawlable, indexed, and the retrieval crawlers allowed.
- 2. Can AI read you? Real text in the HTML, in self-contained chunks it can lift.
- 3. Can AI understand you? What you do, where, and for whom, stated plainly and consistently.
- 4. Can AI trust you? Reviews, local presence, and what other sites say about you.
- 5. Does AI choose you? Named in the answer, not just linked at the bottom.
Here is the uncomfortable pattern: most businesses fail at signals one and two, and spend all of their money on signal five.
Signal 1: Can AI find you?
An engine can only recommend a business it is allowed to fetch. That means your site is crawlable, indexed, and open to the retrieval crawlers specifically. There are two kinds of AI bot, they do different jobs, and blocking the wrong one is a quiet way to disappear from AI answers entirely.
The split is simpler than it looks. Training crawlers collect content to train future models. GPTBot, ClaudeBot and Google-Extended sit in this group. You can block all three and it will not remove you from any search result or AI answer, because none of them are involved in answering live questions.
Retrieval crawlersare the ones that go and fetch pages when someone asks a question right now. OAI-SearchBot for ChatGPT search, PerplexityBot for Perplexity, Claude-SearchBot for Claude's web search, and Googlebot for Google's AI features. These must stay allowed. OpenAI states plainly that sites opting out of OAI-SearchBot will not be shown in ChatGPT search answers, and Anthropic documents its search crawler separately from its training crawler for the same reason.
Sources: OpenAI bots documentation, Anthropic crawler documentation.
The clean setup:block the training crawlers if you have a reason to, allow every retrieval crawler, and never block by pattern. A blanket rule aimed at "AI bots" usually catches the search bots too.
What to do:
- Open your robots.txt and read it line by line, looking for anything that blocks OAI-SearchBot, PerplexityBot, Claude-SearchBot, or Googlebot.
- Check that your key pages are actually indexed, not just live. If Google cannot see them, most AI features cannot either.
- If your site sits behind a firewall, bot filter or aggressive security plugin, confirm it is not silently rejecting these crawlers.
Signal 2: Can AI read you?
AI reads text, not pictures and not code it has to run first. If your services, prices or answers live inside an image, a video, or a script that loads after the page arrives, the crawler sees an empty room. Content is also retrieved in chunks, so short self-contained passages travel much better than long build-ups.
Two things break this signal more than anything else.
The first is JavaScript. A Vercel and MERJ analysis of over 500 million GPTBot fetches found no evidence that it executes JavaScript at all. Client-side rendered content is therefore effectively invisible to AI crawlers. If your page is a shell that fills itself in after load, what the crawler stores is the shell. The same goes for text baked into an image or a PDF that nothing links to in plain HTML.
Source: Vercel and MERJ analysis of GPTBot fetches.
The second is structure. Engines do not read a page the way a person does, front to back. They retrieve pieces of it. One analysis of citation behaviour found that 44.2% of ChatGPT citations came from the first 30% of a page, which tells you two useful things: put the answer near the top, and make each section make sense on its own. A paragraph that only works if you read the three above it is a paragraph that cannot be quoted.
Source: Lumar on content chunking and AI extractability.
What to do:
- View the page source and search for your most important sentence. If it is not there, the crawler cannot see it either.
- Move the direct answer to the top of every page and every section, then explain underneath.
- Write sections that stand alone, with the subject named rather than referred to as "it" or "this".
- Put prices, service names, hours and service areas in real text, never only in a graphic.
Signal 3: Can AI understand you?
Reading you is not the same as understanding you. An engine needs to know exactly what you do, where you do it, and who you do it for, stated in plain text and identical everywhere it looks. When the details conflict, the engine hedges or quietly picks a competitor it is more certain about.
This is the entity clarity signal. Engines build a picture of your business by cross-checking the many places you appear, and confidence comes from agreement. A clinic that describes itself as a "wellness sanctuary" on its homepage, a "med spa" on Google, and a "day spa" on Yelp has handed the engine three businesses instead of one.
The same applies to your name, address and phone details. Inconsistent listings are a long-standing local search problem and they matter just as much now, because an engine that cannot resolve which record is correct has an easy alternative: name someone else.
Source: BrightLocal on name, address and phone consistency.
Structured data belongs here rather than anywhere else on this page. LocalBusiness and Service schema state your identity in a form machines read without ambiguity, which genuinely helps with understanding. What it does not do is make an engine recommend you. Position it honestly in your own head: schema helps machines understand you, it is not the thing that gets you chosen.
What to do:
- Write one plain sentence describing what you do, for whom, and where, then use that same wording everywhere.
- Audit your name, address and phone details across your site, Google Business Profile, and every directory you appear in.
- Name your services the way customers name them, not the way your industry does internally.
- Add LocalBusiness schema as a clarity layer, with realistic expectations about what it delivers.
Signal 4: Can AI trust you?
This is the signal most people get wrong. Trust does not come from your website. It comes from what the rest of the web says about you, and the strongest available evidence points squarely at reviews and local search presence. Technical polish, by contrast, shows very little relationship with whether AI recommends you.
Start with reviews, because the numbers here are hard to ignore. A study of 10,000 local businesses found that businesses visible in both ChatGPT and Perplexity averaged 133 Google reviews, against 11 for the businesses that were invisible. Presence in Google's Local Pack correlated with 3.5 times more AI recommendations. Meanwhile llms.txt files, FAQ schema, Core Web Vitals and page speed showed weak or no correlation at all.
Treat that as directional rather than settled. It is a single vendor-sourced study, and vendor research deserves a raised eyebrow. Worth noting, though: the null findings cut against the vendor's own commercial interest, since the things it found did not work are exactly the things the industry sells. That makes those particular results harder to dismiss.
Source: Study of 10,000 local businesses (vendor-sourced).
Two other findings point the same direction. Ahrefs found that off-site brand mentions correlate with AI visibility more strongly than backlinks do, which fits how these systems work: they are reading text about you, not counting links to you. And across engines, Reddit, YouTube and Wikipedia are among the most-cited domains, meaning a large share of what an engine reads about your category was not written by any business in it.
Sources: Ahrefs on AI brand visibility correlations, Search Engine Land on most-cited domains.
The uncomfortable part:your reviews and your Google Business Profile almost certainly matter more to AI visibility than your website's code does. A slow site with 200 detailed recent reviews tends to be in a stronger position than a fast, beautifully built one with nine. That is not an argument for neglecting your site. It is an argument for where the next hour of effort should go.
What to do:
- Build a steady, ongoing review habit rather than a one-off push, and ask customers to mention the specific service they came for.
- Claim and fully complete your Google Business Profile, including services, hours, photos and categories.
- Get mentioned off-site: local roundups, partner sites, community threads, podcasts, anywhere your category gets discussed.
- Take part honestly where your service comes up on Reddit and YouTube, without pretending to be a customer.
Signal 5: Does AI actually choose you?
Being cited is not the same as being named. An engine can pull a fact from your page, use it in the answer, and never say who you are. The reader gets the information and you get nothing. Being chosen means your brand name appears in the sentence, and some kinds of content earn that far more often than others.
Semrush looked at this directly and found that 61.7% of citations were link-only, with no brand name anywhere in the answer text. That is the gap between showing up in the source list and actually being recommended. The same research found that comparison and listicle content earned brand mentions 2.4 times more often than plain informational content.
The reason is intuitive once you see it. When someone asks "what is the difference between X and Y" or "who are the best options for Z", the answer format itself requires naming names. Informational content answers a question and the engine can absorb it silently. Comparison content cannot be summarised without the brands in it.
Source: Semrush ghost citations study.
What to do:
- Publish comparison-style content: options compared, approaches weighed, honest pros and cons including where you are not the right fit.
- Name yourself clearly in your own content instead of writing entirely in the first person.
- Test real prompts your customers would type, and record which ones name you against which merely link you.
- Retest every few months, because engine behaviour shifts and last quarter's result is not this quarter's.
Where most businesses actually fail
Almost everyone spends their money on signal five while quietly failing signals one and two. They commission content designed to get chosen, on a site the retrieval crawlers cannot reach or cannot read. The chain breaks at the first link, so nothing downstream of it can work.
Fix them in order, because the order is not decorative. There is no point writing a brilliant comparison page if the crawler never fetches it. There is no point building reviews if your business details contradict each other across the web. Each signal only pays off when the ones before it are clear.
The good news is that the early signals are usually the cheapest to fix. A robots.txt line and a rendering change often cost an afternoon. The later signals take longer, but they are also the ones that compound, and most of your competitors have not started on either end.
Frequently asked
What does AI look at when recommending a business?
Underneath the differences between engines, they are all working through the same five questions: can I find you, can I read you, can I understand you, can I trust you, and should I choose you over the alternatives. The first two are technical, the middle one is about clarity, and the last two come mostly from what the rest of the web says about you rather than from your own website.
Does schema markup help with AI search?
Schema markup helps machines understand you. It states your name, services, location and hours in a form that leaves no room for guessing, which supports the understanding signal. What it does not do is make an engine recommend you. In the studies we have seen, reviews and local search presence correlate far more strongly with AI visibility than schema does, so treat schema as good hygiene rather than as the lever.
Do reviews affect AI recommendations?
They appear to matter more than almost anything else. In a study of 10,000 local businesses, the ones visible in both ChatGPT and Perplexity averaged 133 Google reviews, while the invisible ones averaged 11. Reviews are third-party evidence written in the language customers actually use, which is exactly the kind of material an engine can quote with confidence. Volume, recency and specific detail all help.
Why is my competitor showing up in AI and not me?
Usually because they clear a signal you do not. Common causes are a site the retrieval crawlers cannot read, content that only appears after JavaScript runs, business details that conflict across listings, far fewer reviews, or simply less presence in the places engines cite. Work through the five signals in order rather than guessing, because the gap is often earlier in the chain than people expect.
Should I block AI crawlers?
It depends which ones. Training crawlers like GPTBot, ClaudeBot and Google-Extended collect content to train future models, and blocking them does not remove you from AI answers. Retrieval crawlers like OAI-SearchBot, PerplexityBot, Claude-SearchBot and Googlebot are what fetch pages to answer live questions, and blocking those can make you disappear from AI results. Block training if you want to, keep retrieval allowed.
Which of the five signals is stopping you?
You now know what AI checks. The next question is which link in your own chain is broken, and it is rarely the one people assume. The AI Search Audit walks all five signals for your business and tells you, in order, exactly where the break is and what to do about it.
Get the AI Search Audit