How to tell if AI tools cite you
Ask ChatGPT, Gemini and Perplexity what your customers really ask, and write down what comes back.
Arnaud7 min
Your customers see your site. Google and AI read its code. Here's what to fix first.
ArnaudNovember 12, 20268 min
The key points
Summarize withChatGPTClaudePerplexity
Your site has two kinds of visitors. Your customers look at a photo, a headline, a button. Crawlers don't look at anything. A crawler is a program that travels the web on behalf of Google or an AI tool. It receives the page's code, the hidden text that tells the browser what to display, and builds a summary of you from it.
Since July 22, 2026, Google answers on its own in France too, with AI Overviews and AI Mode. ChatGPT, Gemini, Claude and Perplexity do the same. If your code doesn't say clearly what you do, they'll name someone else.
Let's take a made-up example: the website of a roofer in Lyon. Its code is full of tags, small labels that say what each piece of text is for.
What the crawler reads, in excerpts
| The element | Its value on this site |
|---|---|
| The blue title on Google | Roofer in Lyon, roof repair and renovation |
| The page title | Your roof fixed before the next rain |
| The text under the link | Leaks, broken tiles, gutters: we work in Lyon and nearby towns. Free quote, same-day emergency call-outs. |
| What the image describes | A roofer replacing tiles on a roof, above the rooftops of Lyon |
| The business ID card | RoofingContractor, the type meant for a roofer |
The customer sees a photo and a promise. The crawler reads a trade, a town, services and a type of business. But before it can read anything, it has to get in.
Every site can have a small public file, robots.txt, that tells crawlers where they're allowed to go. One extra line is enough to turn them away. Google reminds site owners that your hosting provider, or the service that protects your site, also has to let them through. None of this shows on screen.
These crawlers don't all do the same job. Some read your site to answer a question. Others collect pages to train an AI.
Who reads your site, and why
| The tool | Reads to answer | Collects to train AI |
|---|---|---|
| Googlebot | Google-Extended, a setting more than a crawler | |
| ChatGPT | OAI-SearchBot | GPTBot |
| Claude | Claude-SearchBot | ClaudeBot |
| Perplexity | PerplexityBot | None, according to Perplexity |
ChatGPT, Claude and Perplexity also have a crawler that opens a page on demand, during a conversation. In short: let those through, along with the ones that read to answer. The ones that collect to train AI are your call.
On ChatGPT and Claude, you can refuse training and stay visible in answers. Blocking the wrong crawler is costly: on ChatGPT, a site that blocks OAI-SearchBot no longer appears in its search answers, except as a plain link. So blocking everything at once also takes you out of the answers. Google, OpenAI, Anthropic and Perplexity each publish a list of their crawlers.
Google has added a separate setting. In Search Console, its free tool for site owners, open Settings, then Search generative AI. Include is the default. Exclude removes your site from these answers without touching the rest of your ranking. Unless you have a specific reason, leave it on Include.
Some sites send an almost empty page, then fill it at the last moment with JavaScript, a language that runs in the browser. Your customer sees no difference. A crawler may get an empty page.
Google can run that program as long as it isn't blocked, even if it takes more work. With AI tools, nothing is guaranteed. In a test published by RESONEO in January 2026, ChatGPT, Claude and Gemini read only the text present from the start, while Copilot, Microsoft's AI, ran the program. The test only covers pages read on demand, during a conversation, so treat it as a warning, not a rule.
Google reads what your code says,not what your page shows
Five elements do most of the work. None of them guarantees a ranking.
Google picks it from several sources, and the title tag comes first on its list. Write one per page, with what you do and where. Avoid “Home” and piles of keywords. Our roofer says what and where in one line.
This is the big headline visitors read when they arrive, the h1 tag. Our roofer uses it to speak to customers: “Your roof fixed before the next rain.” The blue title, meanwhile, gives the trade and the town.
You suggest it. Google decides. It starts mainly from the page content, and uses your description if that describes the page better. Write one per page: what the customer will find there, in a few short sentences.
This is the alt text, written in the code and invisible on screen. For Google, it's the most important attribute for describing an image. It also helps screen readers, which read the page aloud for people with low vision. Describe what you see, without piling on keywords.
This is structured data: information organized for machines, invisible on screen. Our roofer states that it's a roofing business, with its address and phone number. Google writes that it isn't required for its AI answers, but recommends it because it makes a page eligible for rich results, listings that show more than a plain blue link. It has to match the visible text on the page.
In the RESONEO test, AI tools didn't read it when opening a page on demand. So also write your contact details, hours and service areas in plain text on the page.
Five well-written elements save machines from guessing.
Perfect tags on a thin page are useless. Google writes that unique, useful content will likely matter more, in the long run, than any other advice in its guide.
Write what only you can say: your real lead times, your service areas, what you refuse to do and why, the questions people ask you on the phone. And keep what matters in text: a price written on a photo is hard to read.
Google and AI tools don't stop at your pages. Your Google Business Profile, not to be confused with the structured data in your code, carries your hours, phone number and reviews. It can make you more visible in Google's AI answers as well as its results. Keep the same hours, phone number and address there as on your site. The rest, what others say about you, is in SEO and GEO, the difference in plain English.
All you need is your browser, on a computer.
Five checks, nothing to install
What Google knows about you
Type site:yoursite.com into Google. You'll see a sample of the pages it has stored. If an important page is missing, dig in.
The door
Open yoursite.com/robots.txt. Look for a Disallow: / line with the slash alone, under User-agent: * or under a crawler's name. It closes the whole site. A path after the slash closes only that part.
The text
Open a page's source code and search for one of your sentences. If it isn't there, ask your provider why.
Your titles
Go back to the site:yoursite.com search and read your titles. Several pages with the same title need rewriting.
Your images
In the source code, search for alt=. If it's empty on a photo that matters, write its description.
Want a second pair of eyes? Our online analysis gives you a first look in a few minutes. The full audit is done by a person: your site page by page, with fixes ranked by priority.

Arnaud
Managing director
Arnaud designs your websites, from experience to interface, and their visibility on Google and AI.
Ask ChatGPT, Gemini and Perplexity what your customers really ask, and write down what comes back.
Arnaud7 min
Nobody sees it, but Google reads it: who you are, what you do, where. What it holds, what it doesn't do, and how to check yours.
Arnaud6 min
Anyone can generate a website. Securing it is a job: someone on our team reviews everything before every launch.
Frédéric8 min