How to Check if Your Website Shows Up in ChatGPT

There is a moment in almost every client call now where someone pulls out their phone, opens ChatGPT, types "best [their service] in [their town]," and watches their competitor get recommended instead of them. It is a gut punch, and it is also the cleanest motivation I have ever seen for fixing a website. The trouble is that one prompt on one day is not a reliable test. Ask the same question twice and you can get two different lists, because the model is sampling, and because whether it browsed the live web that time changes everything.

So before you panic or celebrate, you need a method. Below is the test I actually run, what each result means, and the honest part most agencies skip: some of whether you appear is genuinely out of your hands. I will tell you which parts you can move and which you cannot.

What does it actually mean to "show up in ChatGPT"?

Showing up in ChatGPT means the model either remembers your business from its training data or pulls your page into a live answer and cites it. Those are two separate systems. A business can be remembered but never browsed, or browsed and cited but otherwise unknown. Testing both tells you which lever to pull.

When ChatGPT answers without browsing, it is reciting what it absorbed during training, which is a frozen snapshot taken months earlier. If your business shows up there, it means you were established and described clearly enough on the open web at the time that snapshot was taken that the model encoded you as a real entity. That is a strong signal and a slow one to earn.

When ChatGPT browses, a different thing happens. It runs a search behind the scenes, usually through Bing's index, fetches a few pages, and writes an answer from what it just read, often with little source citations. Here you are not competing on memory, you are competing on retrieval: can your page be found, read, and judged relevant in the half second the model spends looking? These are different races with different finish lines, and a complete check runs both.

The five step test you can run today

Write down five queries a real customer would type, ask each in ChatGPT with browsing off, then ask the same five with browsing on, and record whether you appear and what source got cited. Repeat in Perplexity and Gemini so you are not reading one model's quirk as the whole picture. Keep the results in a simple table you can rerun next month.

Start with the queries, because this is where most people go wrong. Do not search your business name, because a model will usually find a business it is handed by name. Search the way a stranger would when they do not know you exist yet. For a Cassville web developer that looks like:

• "Who builds custom websites for small businesses in southwest Missouri?" — the discovery query, no name, pure intent.
• "I need an SDVOSB web developer for a small federal contract, who should I look at?" — a niche query where a clear specialty should win.
• "What's the difference between a website builder and a hand coded site, and who does the custom version?" — an informational query that can pull your blog in.
• "Recommend someone who does AI search optimization for local businesses." — the service query that tests whether your positioning landed.
• "Best web design companies near Bentonville Arkansas" — the local-pack-style query AI assistants increasingly answer directly.

Now run each one twice. First with the browsing or search tool turned off, so you are reading pure memory. Then with browsing on, so you are reading live retrieval. In ChatGPT, the "Search" toggle controls this; in Perplexity, browsing is always on and it shows numbered sources, which makes it the single best tool for this test because you can see exactly which page it read. Record three things per run: did you appear, in what position, and what URL did it cite. That last column is the gold, because it tells you which of your pages the AI trusts and which competitor pages are beating you.

You did not appear. What does that actually mean?

Not appearing almost never means you are penalized. It means your site was not in the small set of sources the model pulled from for that question. Either you were not encoded into the training data, or your live page lost the retrieval race to clearer competitor pages. Both are sourcing problems, and sourcing problems are fixable.

The mistake is reading absence as a verdict on quality. ChatGPT is not judging that your work is worse; for a discovery query it pulls maybe three to eight sources out of millions of candidate pages, and if yours was not among them, you simply were not in the room when the answer was written. A great business with a thin, vague website loses to a mediocre business with a clear, well structured one, because the second one was easier to find and easier to summarize.

Look at what got cited instead. If it cited a directory like Clutch or a "top 10 web developers" listicle, the path forward is being listed and reviewed on those third-party pages, not just polishing your own site. If it cited a competitor's service page, read that page: it almost certainly states plainly what they do, where, and for whom, in language a model can lift in one sentence. If it cited nothing relevant and gave a generic answer, you are in an underserved query where being the clearest specific answer is a wide-open opportunity. The cited source is your map of where to work.

What actually moves the needle on AI visibility

Five levers, in rough order of impact: be present on the pages AI already trusts, define your business as a clear entity, mark up your pages with schema, accumulate real reviews, and earn third-party mentions. None of these is a trick. They are the same signals that make a model confident enough to say your name out loud.

The first lever is presence on trusted sources. Models lean heavily on a handful of authoritative places — Wikipedia and Wikidata, established industry directories, news mentions, and a few large review platforms. You do not need all of them, but being absent from every one of them is why a business is invisible to the no-name discovery query. The second lever is entity clarity: your site should state, in plain declarative sentences, exactly who you are, what you do, where you operate, and who you serve. Models extract entities from sentences like "ThatDeveloperGuy is a service-disabled veteran-owned studio in Cassville, Missouri that hand-codes websites for small businesses." Vague hero taglines give them nothing to grab.

The third lever is structured data. Schema markup — Organization, LocalBusiness, FAQPage, Article — hands the machine the facts in a format it cannot misread, which is most of what my AEO services work tightens. The fourth is reviews: a steady stream of genuine reviews on Google and the platforms in your industry is one of the strongest "this business is real and respected" signals a model can read. The fifth is third-party mentions — a guest article, a podcast, a local news piece, a partner site linking to you — because a model trusts a business other sites talk about far more than one that only talks about itself. When I wrote the companion piece on how to get your business into ChatGPT search results, every tactic in it ladders up to one of these five.

The honest part: how much of this can you control?

You control your website, your schema, your reviews, and your effort to earn mentions. You do not control the model's training cutoff, its sampling randomness, which sources it weights, or whether it browses on a given query. Roughly, you can strongly influence the live-retrieval game and only slowly influence the from-memory game. Anyone promising guaranteed ChatGPT placement is selling something that does not exist.

This is the part agencies tend to skip, so here it is plainly. The from-memory layer moves on the model maker's schedule. If OpenAI trained a snapshot in January, nothing you publish in June exists in that memory until the next training run, which could be many months out and is entirely their call. You can earn your way into the next snapshot by becoming well-described across the open web, but there is no button and no timeline you control. That is genuinely out of your hands, and you should be suspicious of anyone who claims otherwise.

The live-retrieval layer is where your effort pays off fastest, because a page you publish today can be browsed and cited tomorrow. There is also irreducible noise: ask the same question twice and the answer varies, which is why you test five queries across three tools and watch the trend over months instead of reacting to one run. The realistic goal is not "ChatGPT always recommends me first." It is "across many askings of the questions my customers actually use, I show up materially more often than I did last quarter, and when I do, it cites my page and gets the facts right." That is a winnable game, and it overlaps almost entirely with the generative engine optimization work that makes you citable in the first place.

How often to retest, and what to log

Test the same five queries once a month, in the same three tools, and log appearance, position, and cited source each time. The single run is noise; the multi-month trend is signal. A simple spreadsheet beats any paid "AI rank tracker" for a small business, because you can see exactly what changed and why.

Consistency is the whole point. Because answers vary run to run, a one-off check in March versus a one-off check in May tells you almost nothing. The same five queries, asked the same way, on the first of each month, gives you a trend line. If you fixed your schema and earned two new reviews in April and your appearance rate climbs in May, you have actual evidence that your work moved something, which is rare and valuable in this corner of marketing.

Keep it boring: a spreadsheet with a row per query and a column per month, noting whether you appeared and which URL got cited. Watch the cited-source column especially — when a competitor's page that kept beating you stops getting cited and yours starts, that is the day your work landed. You do not need a tool subscription for this. You need fifteen minutes and the discipline to do it the same way every month, which is exactly the cadence I build into ongoing client work.

Related Internal Links

Use these to go from testing your visibility to actually improving it across AI assistants.

FAQ