Tracking What Le Chat Says About Your Brand

Le Chat is Mistral's consumer AI assistant, and tracking it starts with an honest gap. The evidence base behind this library holds verified crawler and retrieval documentation for OpenAI, Google, Anthropic, Perplexity, Meta and Microsoft. It holds nothing first-party from Mistral.

As of October 2026, that means this page can't confirm:

• which user-agent strings Mistral's crawlers send
• whether those crawlers obey robots.txt, or which fetches are user-triggered
• when Le Chat decides to search the web, and whether that's a user toggle or the model's choice
• which search index, if any, sits behind its answers

Read Mistral's own crawler and product documentation before you write a robots.txt rule for it. Don't copy a user-agent name from a post that cites no source.

Other vendors show why the details matter. OpenAI splits its bots by job: GPTBot governs training, and OAI-SearchBot governs ChatGPT search. OpenAI states that "Each setting is independent of the others." If Mistral documents a similar split, the same logic applies. Blocking a training bot and blocking a search bot have different consequences, and a blanket AI block removes both.

Every AI answer comes from one of two places. It either comes from the model's frozen training data (the parametric stage), or from pages fetched at answer time (retrieval). Nothing you publish this quarter reaches a parametric answer. A retrieval answer can draw on your current pages, if the crawler could reach them.

On every platform the evidence base documents, retrieval fires per query and at someone's discretion. ChatGPT search is model-discretionary. Claude.ai search is a user toggle with no published default. Le Chat's behavior is undocumented in our evidence base, so treat it the same way: unknown until you observe it.

In practice:

• Answer shows links or sources. A web fetch probably happened. Your current site could have shaped it.
• Answer shows no sources. You're probably reading the model's memory. Fixing your site won't change it soon.

"Probably" is the honest word. A citation also isn't proof of use. OpenAI exposes retrieved and cited URLs as separate lists, and the retrieved list is larger. A 2026 mechanistic study (arXiv:2606.28358) showed citation output can be changed without changing the sources. Count what Le Chat cites, but don't read it as a full record of what it used.

A spreadsheet and a fixed routine beat ad hoc checking. Here's the method.

1. Write 10 to 20 buyer questions. Mix category questions ("best payroll tool for a 20-person agency"), comparisons, and direct fact questions about you (price, what you do, where you're based). Decide early whether to include prompts that name your competitors.
2. Keep the session clean. Other platforms let memory reshape answers (OpenAI documents memory shaping ChatGPT's search queries). Use a fresh session with no custom instructions so runs stay comparable.
3. Note the search setting if Le Chat shows one, every time.
4. Log every answer with the fields below.
5. Rerun on a fixed schedule with the same wording. Change one thing at a time.

Say you run a hypothetical accounting firm in Leeds. If Le Chat names you in a sourced answer but quotes last year's fee, that row is a fix target. If it never names you and cites three directory sites, that row points somewhere else entirely.

If Le Chat never cites your site, check whether its crawlers can reach the site at all. These checks have the strongest evidence of anything in AI visibility, because they remove blocks.

• robots.txt is per bot. Rules apply to named user agents (the Robots Exclusion Protocol, RFC 9309). Confirm which Mistral agents Mistral documents, then check your file names them the way you intend.
• A firewall or CDN can override robots.txt. OpenAI's help page says ChatGPT search eligibility needs the bot allowed and the host or CDN letting its IP addresses through. The same mechanism can block any crawler. From 2026-09-15, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains, new sites on existing accounts, and all existing free-tier customers. You may have lost access without editing anything.
• JavaScript-only content may be invisible. Vercel found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. A June 2026 re-test confirmed this across GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and others. No Mistral crawler was in that set, so its behavior is untested. The safe move: put prices, product names and a one-line description in the raw HTML.

99Visibility's AI Crawler Checker requests a page as each crawler it covers and shows which ones your firewall turns away. Snippet controls like nosnippet are documented by Google and Bing. Nothing in the evidence base says Mistral honors them.

Most rows will say "named" or "not named". The useful rows are more specific.

• Wrong facts. A stale price or a defunct product name is concrete and fixable. State the correct fact plainly on a page crawlers can fetch. That's the only on-page change with a positive experimental signal: it gives the engine a sentence it can quote. For automated flagging, see tools that alert you to false AI claims.
• Third-party sources. When Le Chat cites review sites or directories instead of you, being named on those sites is associated with appearing in the answer. It's an association, not a promise.
• Suspicious sources. Hidden text and prompt injection on third-party pages can skew AI answers. Read whether a competitor can manipulate AI answers. Never use these tactics yourself. They're illegitimate, and engines treat them as attacks.

Don't report a rank. No engine documented in the evidence base exposes answer position, so there's nothing to check a rank against. Keep Le Chat in its own column too. Blending it with ChatGPT brand monitoring or Claude's answers into one number hides which engine has the problem.

You control access: robots.txt rules, firewall settings, server-rendered HTML, and the accuracy of facts on your own pages. You can also earn mentions on the third-party sites engines cite.

You don't control Le Chat's training data, its retrieval decisions, or which sources it picks. A critical survey of 45 GEO studies (arXiv:2607.14035, July 2026) found no technique with a stable, cross-platform causal effect on discoverability. Anyone promising Le Chat citations is promising something nobody has measured.

For tooling, 99Visibility's paid plans audit ChatGPT, Google AI Overviews, Gemini, Perplexity and Claude. Le Chat isn't on that list, so the manual log is the practical tool for it today. The same approach works for other engines with thin documentation, like DeepSeek. For engines that are covered, compare options in free AI brand monitoring.

Does Le Chat use the same crawlers as ChatGPT? No evidence says so. GPTBot, OAI-SearchBot and ChatGPT-User are OpenAI's crawlers, documented on OpenAI's bots page. robots.txt rules apply per named user agent, so allowing an OpenAI bot does nothing for any other company's crawler. Our evidence base holds no Mistral crawler documentation as of October 2026. Check Mistral's own pages for current agent names before you edit robots.txt.

Will schema markup or an llms.txt file help Le Chat cite me? Nothing supports it. No engine in our evidence base documents reading llms.txt, and Google says Search ignores it. Ahrefs' matched test found AI Overviews citations fell 4.6% after pages added schema. Nobody has tested either on Le Chat. Existing markup does no harm, but it isn't a lever for AI answers. Spend the time on access checks and correct facts instead.

How often should I rerun my Le Chat prompts? Pick a fixed rhythm, weekly or monthly, and keep the prompt wording identical so changes mean something. Models update and retrieval behavior shifts without notice, so one run is only a snapshot. Rerun sooner after you fix something on your site, such as moving key facts into the raw HTML or changing firewall rules, and note the fix date in your log.

Can I get a single Le Chat visibility score? You can count rows, such as the share of prompts where Le Chat names you. Treat that as your own tally, not a measurement Mistral validates. Any single score averages answers given with and without live retrieval, and Le Chat's retrieval rules are undocumented in our evidence base. Report named-or-not and wrong facts per prompt, with search state recorded, rather than one blended number.