Which AI visibility tools track Claude as well as ChatGPT, Gemini and Perplexity?

Plenty of trackers put Claude on their platform list. 99Visibility's paid plans, for example, audit Claude alongside ChatGPT, Google AI Overviews, Gemini and Perplexity. The list is the easy part.

The hard part is that Claude does not give one answer to a buyer question. It gives at least two kinds:

• Search off. Claude answers from what the model learned in training. Nothing on your site today can change that answer.
• Search on. Claude may fetch pages, read them and cite them. Your site can now enter the answer, if Claude's crawlers can reach it.

A tool that only queries one way reports one of these. A tool that mixes them reports an average of both. Neither is wrong, but you need to know which one you're buying. Our guide to tracking Claude's answers covers the method in more depth.

Three facts make Claude a moving target, as of October 2026.

The consumer app leaves search to the user

On claude.ai, web search is a toggle. Anthropic publishes no default for it. So two buyers asking the same question can get a search-grounded answer and an ungrounded one.

The model decides when to search

Even with search available, Claude chooses whether to run it. On the API, the integrator's system prompt also steers that choice. A tracker's own prompt wrapper can push results one way or the other.

The API is not the app

Anthropic's API web search tool is enabled at the organisation level by default, but that setting does not describe consumer behaviour on claude.ai. A vendor calling the API is measuring a proxy for the app. That gap exists for every engine, and our comparison of API answers vs consumer app answers explains where it bites.

What this means: any Claude score is a choice the vendor made. Your next move is to make them tell you what it was.

With search off, the answer reflects the model's training data and its knowledge cutoff. Citations don't exist, because no pages were retrieved. If a tool reports a Claude citation rate on search-off answers, that metric is empty.

With search on, Claude's API returns each result's URL, title and a page_age field, plus citations tied to specific text spans of up to 150 characters. That makes Claude citations precise, but it also makes them a different unit from Perplexity's, which cites whole sources by number. Adding the two together produces a figure with no shared denominator.

Citation is also not the same as use. Claude can read a page and not cite it. A tool that only counts citations undercounts influence.

If you want one view across engines, read whether one dashboard can track every AI engine first, and insist on per-engine numbers underneath any blended score.

Use these on every tool you evaluate, including ours.

1. Do you query Claude with web search on, off, or both? If both, are the results reported separately?
2. Which Claude model do you call, and when did it last change? Model swaps move results without anything changing on your site.
3. Do you call the API or measure the consumer app? If the API, what system prompt wraps your queries?
4. How do you count a Claude citation? Span-level citations should not be added to whole-source citations from other engines without saying so.
5. Do you record what was retrieved but not cited? Anthropic exposes retrieved results, so this is measurable.
6. Do you check factual accuracy, or only mentions? A mention with a wrong price is worse than no mention.
7. How many prompts back each number, and how often do you re-run them? A small sample swings week to week.

Price matters too. If budget drives the decision, see the cheapest tool that covers all four engines, then hold the cheap option to the same seven questions.

No tool and no tactic decides what Claude says. What you control is whether Claude's search can reach and read your pages.

• Robots.txt is per bot. Anthropic runs ClaudeBot (training), Claude-SearchBot (search quality) and Claude-User (fetches a user triggers). Blocking ClaudeBot is a training decision. It is not the search control.
• Firewalls can override robots.txt. A CDN or WAF can turn away a crawler your robots.txt allows. Anthropic warns that IP blocking also stops it reading your robots.txt at all. The free AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
• Render the facts in HTML. Vercel found Claude's crawlers fetched JavaScript files on 23.84% of requests but did not execute them. Prices, product names and definitions that appear only after JavaScript runs may be invisible.

Fix those blocks first. Then measure, with search state disclosed, as described in our guide to LLM visibility tracking.