The AI models we track
Ask two AI models the same question about your category. You can get two different shortlists.
That's not a glitch. Some models search the web first and show you every search. Some put products in front of the buyer before a single link. One doesn't always answer at all.
Here's what each model gives you to track, and the one thing to check on your site for each.
What each model shows you
Every model on your plan gets the same prompts on the same day.
What comes back isn't the same. Here's the difference at a glance.
| Model | Collected from | Shows its query fan-out | Sources | Shopping results | Answers every prompt | |
|---|---|---|---|---|---|---|
| Citations | Retrieved sources | |||||
| App | Yes | Yes | Yes | Yes | Yes | |
| API | Yes | Yes | Yes | No | Yes | |
| App | No | Yes | Yes | Yes | No | |
| App | No | Yes | Yes | Yes | Yes | |
| App | Yes | Yes | No | Yes | Yes | |
| App | Yes | Yes | Yes | No | Yes | |
| App | No | Yes | Yes | No | Yes | |
ChatGPT
Made by OpenAI. ChatGPT writes an answer and decides for itself whether to search the web first. When it does, you see every search it ran.
It reads far more than it links.
Each answer lists its retrieved sources: pages it pulled in but didn't cite. Those are the pages it looked at and passed over. If yours is one of them, you were close.
Some answers put product cards in front of the buyer, with prices and shops. Those count in your shopping results.
The pages ChatGPT reads and skips are your shortlist of fixes.
- Let it in: OpenAI's search crawler is
OAI-SearchBot. Check your robots.txt doesn't block it, and that your CDN lets OpenAI's published IP addresses through.
The whole method, step by step, is in our guide: How to track your visibility in ChatGPT →
Claude
Made by Anthropic. Claude writes an answer and can search the web before it does.
We ask Claude through Anthropic's own API, with web search switched on. We keep three things: the searches it ran, the pages its search found and the pages it cited.
Claude returns no product cards, so it adds nothing to your shopping results. It's available on Enterprise.
Claude cites what its search finds. Make sure its search can find you.
- Let it in: Anthropic names two crawlers behind Claude's answers:
Claude-SearchBotandClaude-User. Blocking either can cut your visibility in Claude.ClaudeBotis the one that collects pages for training.
Google AI Overviews
Made by Google. An AI Overview is the short answer above the results of an ordinary Google search. The buyer doesn't choose to ask an AI. It's just there.
Except when it isn't.
Google doesn't show an overview for every search. The same search can bring one on one day and none the next. We count a day with no overview as a day you weren't named, because a buyer searching that day didn't see you either.
When it does show one, it links its sources beside its sentences and lists more beside the overview. For some shopping searches it adds a row of products. Those count in your shopping results.
An overview Google doesn't show can't name you. Count those days. Don't hide them.
- Let it in: AI Overviews uses Google's ordinary crawler,
Googlebot. If your page can appear in Google Search with a snippet, it can appear in an overview. BlockingGoogle-Extendeddoesn't take you out.
Google AI Mode
Made by Google. AI Mode is a separate mode in Google Search. The buyer reaches it through the AI Mode button, or by asking a follow-up in the box under an AI Overview. What they get is a longer, chat-style answer.
AI Overviews appear on their own. AI Mode is asked for. Because the buyer asked, it almost always answers, and its answers run long, often with headings, lists and tables.
On shopping prompts it shows a row of product cards. Those count in your shopping results.
In AI Mode, the product card is often the answer. Track it like one.
- Let it in:
Googlebot, the same as AI Overviews.Google-Extendedmakes no difference here either.
Microsoft Copilot
Made by Microsoft. Copilot searches the web before every answer, usually once and sometimes up to four times. You can see every search it ran.
It links only the pages it cited. There's no list of pages it read and skipped.
So its searches are your best clue to what it went looking for.
Some answers include product cards. We keep them with the answer, but they don't count in your shopping results.
With Copilot, the searches tell you what it was looking for. Make sure your page answers them.
- Let it in: Copilot uses Bing's crawler,
Bingbot. Don't block it. A NOARCHIVE tag on a page keeps that page out of Copilot's answers too.
Perplexity
Perplexity searches the web before almost every answer, then writes a short answer and lists the pages it used. You see every search it ran.
It shows both kinds of page: the ones it cited, and the ones it read and didn't. The second list is the useful one: it shows you where you nearly made it.
It shows no product cards.
Perplexity shows you what it read and skipped. Start there.
- Let it in: Perplexity's search crawler is
PerplexityBot. Perplexity recommends allowing it, and says it isn't used for training.
Gemini
Made by Google. Gemini is Google's AI assistant, separate from Google Search. The buyer opens it, asks a question and gets a written answer.
Gemini doesn't show the searches behind its answers, and it links almost everything it reads. So there's almost no skipped list.
You're either one of the pages it cites, or you're not in the answer at all.
It shows no product cards.
With Gemini, the citations are the whole story.
- Let it in: Gemini has no crawler of its own. A robots.txt rule called
Google-Extendeddecides whether Gemini can use your pages in its answers. Block it and you drop out of Gemini, though not out of Google Search.
How is every model tracked?
We ask every model on your plan your full prompt set once a day, and keep every answer.
Same prompts, same day. So a gap between two models is a real difference in how they answer. Not a difference in when you asked.
Every model can give a different answer to the same prompt from one day to the next. One check is a snapshot. A daily series is a trend. Make decisions on the trend.
We check each answer for your brand and every competitor you track, and for the sources it used. Every score on the dashboard links back to the answer it came from. The full method is in how we measure AI visibility.
How many models you track depends on your plan. Starter, Growth and Scale each track three models of your choice, and Enterprise tracks all seven, including Claude. See pricing.
Tracking by model, answered.
Which AI models does AIS Insights track?
Up to seven, depending on your plan: ChatGPT, Claude, Google AI Overviews, Google AI Mode, Microsoft Copilot, Perplexity and Gemini. Every model on your plan gets the same prompts on the same day, so a gap between two models is a real difference in how they answer.
How many models can I track?
It depends on your plan. Starter, Growth and Scale each track three models of your choice, and Enterprise tracks all seven, including Claude. You can change your choice as your plan allows.
Do all the models get the same prompts?
Yes. One prompt set, asked on every model on your plan, every day. A prompt where ChatGPT names you and Gemini doesn't is a real gap, not a difference in when you asked.
Does a model give the same answer every time?
No. The same prompt can name different brands from one day to the next, on every model we track. That's why a single check is a snapshot and a daily series is a trend.
Why do some models show no searches?
Google AI Overviews, Google AI Mode and Gemini don't show the searches behind their answers, so there's nothing to keep. ChatGPT, Claude, Microsoft Copilot and Perplexity do, and we keep every one.
Can I see AI model traffic in Google Analytics?
Partly. ChatGPT tags its links utm_source=chatgpt.com, so those visits are easy to find. The Google models add no tag of their own, and Search Console counts AI Overviews and AI Mode clicks with your ordinary search clicks. Either way, treat it as the floor. Clicks miss the buyers who read your name in an answer and came back another way.
Which crawlers should I let into my site?
The ones each company says its answers depend on. OpenAI: OAI-SearchBot. Anthropic: Claude-SearchBot and Claude-User. Google AI Overviews and AI Mode: Googlebot. Gemini: don't block Google-Extended. Microsoft Copilot: Bingbot. Perplexity: PerplexityBot. Blocking a training crawler, such as GPTBot or ClaudeBot, doesn't take you out of the answers.
Know what every model says about you.
You won't fix your visibility on seven models this month.
But you can find the one prompt where your buyers' favourite model names someone else. Fix the page it should have found. Check again tomorrow. Then find the next one.
Your first scan is free and runs your full prompt set on every model on your plan.