Table of Contents: How Answer Engines Actually Build a Response · Signal 1: Can the Engine Retrieve and Read Your Page? · Signal 2: Your Citation Footprint on Other Sites · Signal 3: Are You an Unambiguous Entity? · Why the Phrasing of the Question Changes the Answer · How to Become the Answer · How to Measure Whether Any of This Worked · Frequently Asked Questions
An AI mention is when an answer engine such as ChatGPT, Perplexity or Google’s AI Overviews names your business in its answer. A citation is when it links to the page it pulled that answer from. For a growing number of buyers, one of those two is the first time they hear about you.
I’ve run Paul Teitelman SEO Consulting in Toronto for more than 15 years, and the question I hear most from clients right now is some version of “Why does ChatGPT recommend them and not us?” The answer comes down to three signals, and only one of them lives on your own website.
Most of the confusion here comes from a reasonable assumption that turns out to be wrong. People imagine the model was trained on the internet, remembered who the good businesses are, and recites that memory when asked. If that were true, your only hope would be to have been famous before the training cutoff.
That is not how modern answer engines work for the queries that matter to your business. Perplexity searches the web on nearly every question. ChatGPT does it whenever a question has a local, commercial, or time-sensitive angle. Either way, the system runs a live search, reads the pages it gets back, and writes its answer from what it just found. The citations you see underneath are the sources it used.
That changes where your effort should go. You don’t need to have been famous before a training cutoff. You need to be easy to find and easy to quote today, and that’s work you can start this quarter.
Engines rarely search for exactly what the person typed. Google calls its version “query fan-out”: AI Overviews and AI Mode may split one question into several related searches across subtopics, gather results for each, then write a single answer. ChatGPT and Perplexity work in a similar way. One question can trigger a handful of searches, and your page only needs to win one of them to end up in the answer.
It also explains something you’ve probably noticed. Two people asking what looks like the same question can get different businesses recommended, because the searches running behind the scenes went in different directions.
The first filter is mechanical. Before an engine judges whether your content is any good, it has to be able to fetch and read it, and plenty of sites fail this step without anyone knowing.
Answer engines send their own crawlers to fetch your pages, and the one that matters for visibility isn’t always the one people expect. OpenAI runs GPTBot to collect training data and OAI-SearchBot to find pages for ChatGPT search. Blocking GPTBot keeps your content out of training but doesn’t remove you from ChatGPT’s answers. Blocking OAI-SearchBot does. Google’s AI Overviews and AI Mode work from the regular Googlebot index, so the separate Google-Extended setting, which covers Gemini training, has no effect on them.
| Engine | Crawler that decides whether you can be cited | Training crawler you can block without losing citations |
| ChatGPT | OAI-SearchBot | GPTBot |
| Perplexity | PerplexityBot | None separate |
| Google AI Overviews and AI Mode | Googlebot | Google-Extended |
| Claude | Claude-SearchBot | ClaudeBot |
I find blocked search crawlers on client sites regularly, and it’s rarely a decision anyone made on purpose. It’s usually a firewall rule or a security plugin default that quietly serves a challenge page to anything that looks unusual. To check yours, open yourdomain.com/robots.txt and look for the crawler names above, then ask whoever manages your hosting or CDN whether AI bots are allowed through.
The second half of this filter is rendering. Most AI crawlers don’t run JavaScript at all. A study by Vercel and MERJ of hundreds of millions of crawler requests found that OpenAI’s and Anthropic’s crawlers download JavaScript files but don’t execute them, while Googlebot renders the full page. If your service descriptions and prices only appear after scripts run in a browser, those crawlers get a page with nothing worth quoting.
Here’s a two-minute test. Open one of your service pages, press Ctrl+U (Cmd+Option+U on a Mac) to view the source, and search for a sentence from the main content. If it isn’t in the source, an AI crawler probably can’t see it either. Then there is the question of whether your content is quotable once it’s read, because engines lift self-contained fragments rather than whole pages. A paragraph that opens “This approach works because” makes no sense once it’s pulled out of your page, so it does not get pulled. A paragraph that opens by naming the thing it’s about travels well. That is the whole practical logic behind structuring content around entities rather than keyword strings, and that post walks through how to do it.
This part usually surprises business owners, and it often changes where some of the marketing budget goes. When an engine answers “who is the best X in Toronto,” it very often does not primarily quote the businesses themselves. It quotes the pages that discuss those businesses. Roundups, industry publications, directories, forum threads, review sites. Your own website is one voice in that conversation, and it’s the one voice the engine knows is selling something.
So your visibility depends heavily on whether you appear in the sources the engine already trusts for your topic, which has a clear practical consequence for how you spend. Map the citation landscape for your top ten customer questions, note which domains keep appearing, and treat that list as your outreach targets. A placement on one publication the engines actually quote is worth more than a dozen links from sites that never surface. We saw the flip side in our own testing: on research-style questions, engines cited the retailer’s pages more often than they named the retailer. Being used as a source and being recommended aren’t the same thing, and the gap between them is usually filled by what other sites say about you.
This is where AI visibility and link building merge into the same job, with one important difference from classic link building. A mention can work without a link at all, because the engine is reading the text rather than only following the hyperlink graph, which widens what counts as a win.
In search, an entity is a specific thing a machine can identify with confidence, such as a person, a business, or a place, rather than just a string of words. Engines work on probabilities, so when they’re choosing between a business they understand clearly and one they’re unsure about, they name the one they understand. If an engine can’t tell who you’re, it tends to leave you out.
Entity clarity is mostly unglamorous consistency work:
None of this will win you a mention by itself. All of it removes reasons to skip you.
We tested this directly. In September 2026, we ran an experiment on our own brand and on a Toronto furniture retailer we work with.
We took 10 buying questions for each business and wrote every one three ways: a short keyword, a conversational question, and a longer research-style prompt. The intent stayed the same and only the wording changed. Then we ran every version three times, a week apart, through Rankscale across five engines: ChatGPT, Perplexity, Google AI Overviews, Google Gemini, and Google AI Mode.
The pattern was clear. The retailer was named in 35% of answers to the short keyword versions and in 5% of answers to the research-style versions, never in the top three, for the same buying decisions. All five engines ranked the phrasings in the same order: keyword first, conversational second, research last.
Our own brand did no better. We were named in 10% of keyword answers and in none of the research-style answers, on any of the five engines.
| How the question was phrased | Toronto furniture retailer named | Paul Teitelman SEO Consulting named |
| Short keyword (“sectional sofa toronto”) | 35% | 10% |
| Conversational question | 15% | 5% |
| Research-style prompt | 5% (never top three) | 0% |
It’s a small study, 10 questions per business, so treat the exact percentages as directional. The direction held on every engine, though.
So your AI visibility isn’t one number. You can be the obvious answer to “sectional sofa toronto” and be completely absent when someone asks an engine to compare materials, durability, and ten-year cost. Those are the same customer at different moments of the same decision, and the second moment is usually closer to the purchase.
Most businesses have optimized, accidentally, for the shortest version of the question. The deeper, more considered phrasings are wide open.
Taking the three signals together, here is the order of work I would actually follow:
That sequence is what our AI search audits work through, and the order matters. If your site blocks OAI-SearchBot, no amount of content work will get you into ChatGPT’s answers.
Be realistic about timing, too. Unblocking a crawler can show results within weeks; OpenAI says robots.txt changes reach its search systems in about 24 hours. Building presence in the sources engines trust usually takes three to six months of steady work, and it compounds from there.
Everything above is guesswork without measurement, and this is where a lot of businesses get discouraged, usually because they’re checking the wrong thing.
Don’t judge AI visibility by asking ChatGPT a question once. Answers vary between runs, so a single check tells you almost nothing. What works is picking a fixed set of the real questions your customers ask, tracking them weekly across the engines that matter, and recording two things: whether you were named, and who got cited instead. The second column is the more useful one, because it’s your target list. For our own site, we track 100 questions every week in Rankscale, and the “who got cited instead” list is what sets our outreach priorities.
Google is also bringing AI performance metrics into Search Console, which will put first-party impression data next to your third-party tracking. Use both of them together, because one tells you what Google sees while the other tells you what the rest of the ecosystem is saying about you.
One caution applies to the numbers you report internally. If your tracked questions include your own brand name, your visibility score will look flattering and mean nothing, because of course an engine names you when asked about you. Track unbranded questions, the ones a customer asks before they have heard of you, and you will get a number that reflects reality.
Our AI search audit checks the three signals in this article on your own site: whether the search crawlers can reach and read your pages, where you show up in the sources engines already trust, and how clearly they understand who you are. You’ll also see which of your customers’ questions name you today and who gets named instead. It’s a clear starting point, ranked by what to fix first.
Get in touch, and we’ll show you where your business stands in AI search.
Make sure its crawlers can reach and read your pages, write self-contained answers to the questions your customers actually ask, and get discussed on the third-party sites ChatGPT already cites for your topic. Mentions on trusted sources usually make a bigger difference than changes to your own site.
For commercial, local and current questions, large language models (LLMs) mostly rely on live search. The engine finds and reads pages at the moment of the question, which is why recent work on your site and your citation footprint can change your visibility within months rather than years.
Usually one of three reasons: its crawler can’t properly read your site, your competitor appears in more of the third-party sources it trusts, or your business is a less certain entity than theirs. Work through those three in that order, since the cheapest fix sits first.
It helps as a clarity signal rather than a ranking lever. Valid, specific structured data makes it easier for an engine to be confident about what you’re, which reduces the chance it picks a competitor it understands better.
No. GPTBot collects training data. ChatGPT search uses a different crawler, OAI-SearchBot, so that’s the one you need to allow if you want to be cited. You can block GPTBot and allow OAI-SearchBot at the same time.
It depends on what’s holding you back. Fixing a blocked crawler or a page that only loads through JavaScript can show up within weeks. Earning mentions on the sites engines already trust usually takes three to six months of steady work before it shows in your tracking.
Weekly, against a fixed list of unbranded customer questions. Single spot checks vary too much between runs to be meaningful, and monthly checks are too slow to connect a change you made to a result you got.
Key Takeaways A code-based website is built from a Git repository rather than a CMS…
Key Takeaways Ahrefs estimated 14.3K monthly organic visits for Brewtiful Living. The owner's real analytics…
An AI search readiness audit is a structured review of whether AI answer engines like…
Key Takeaways Scaling AI content without governance can create crawlability, indexing, and site-wide quality issues.…
Key Takeaways Google's AI Performance Report shows when your pages appear in AI Overviews and…
Conversion-focused SEO is the practice of optimizing for search visibility and actual business outcomes like…