How to track what Claude says about my company?

What do AI engines answer when asked how to track Claude?

New Map's measurement¹ asked 68 questions about AI visibility tools across Perplexity, Claude Code and Gemini and collected 146 answers. That is one week's reading of 68 questions about AI visibility tools on 3 engines (146 answers), not a trend: Perplexity cited a source in only 7 of its 68 answers, and Claude Code's 10 answers are too few to read alone.

The same measurement covers this exact question, "How to track what Claude says about my company". It drew 5 answers across Gemini, Perplexity and ChatGPT over 2 weekly readings. Peec AI was named in 2 of the 5 answers. Ahrefs, Otterly^{2, 3} and Profound were each named in 1 of 5, and New Map was named in 0 of 5.

Five answers is a small base, so read the names as who showed up, not as a ranking. Our read: the named companies are all tools that measure brand mentions in AI answers, so the engines pointed to measurement software rather than to a way of checking Claude directly. The chart shows how many of the 5 answers named each company. Method: how New Map measures.

How do you set up the questions and runs to track Claude?

Tracking is a routine, not a one-off check. These steps turn it into a weekly log, and the method is explained in how a brand's visibility in AI answers is measured.

1. Write two question sets: what buyers ask before they know you, and what they ask when they check your company by name. Keep the sets apart.
2. Run each question several times. SparkToro's January 2026 study of 2,961 runs found ChatGPT and Google AI repeated the same brand list in under 1 in 100 runs.
3. Log one row per answer: every company named, every domain cited, the date and the engine.
4. Record whether your own URL was cited, and keep a count of those answers per week.
5. If you sell in Germany or France, keep those markets as separate question sets. Otterly's October 2026 study of 3.95M citations found 64.9% of German AI citations carried a German-market signal, and its French study found 78.8% local-site citations once hidden .com French domains were counted.
6. Repeat weekly with the same questions, so each week can be set beside the last.

The list is the whole tracking loop: ask, log, count, repeat.

Which sources should you log when Claude cites them?

Log the sources, not only the brand names. New Map's measurement found Claude Code cited at least one source in 10 of 10 answers, with 73 sources cited in all. Gemini cited at least one source in all 68 answers; Perplexity in 7 of 68.

Claude Code is a different surface from the Claude app, and it had 10 answers against 68 for Perplexity and Gemini. Read its figure as a hint, not as Claude's behaviour.

Engines do not share sources much. Profound and Otterly data from 2026 put the overlap between Claude and ChatGPT cited domains at about 8% to 13%. Otterly's 20 September 2026 reading of 1,600 prompts daily on 7 engines gave Claude a Reddit share of 0, against 5.4% for Google AI Overviews.

Count sources by origin, not by copies. New Map's October 2026 review of crypto AI-visibility research notes that ten syndicated copies of one press release are one claim. The chart shows sources cited per answer in the same reading.

How often should you check Claude, and what counts as a real change?

Peec AI's September 2026 analysis of 120M chats found weekly moves under about 5 points and monthly moves under about 2 can be chance at typical run counts. Profound's September 2026 study of 1.19M page-engine lifecycles found the median page loses half its citation share within 11 days of peak.

Our read: a weekly check with repeated runs fits both findings, and a single run does not. Treat a move as real only when it holds across runs and across weeks.

Keep the loop short so fixes arrive while the old answer still stands. The diagram shows the weekly check from question to fix.

What do you do when Claude says something wrong about your company?

New Map has not measured the sources of Claude's wrong facts in this reading, so the points below are causes to check, not findings.

Start where the fact begins. New Map's October 2026 review of crypto AI-visibility research found aggregators inherit their upstream's facts, so a wrong claim is fixed at its origin and checked against independent origins.

Sometimes there is no page to fix. Chen and Koudas, a preprint covering 158,099 answers, found ChatGPT names brands in answers without citing any supporting page. That study tested ChatGPT, not Claude.

Name confusion is another cause to check. WordLift⁴'s test found Wikidata first-hit lookup succeeded on 288 of 351 marked names. Nearly all of the 61 failures involved a famous meaning of the name ranked above the correct entity.

Keep your own pages consistent. Gartner, cited by Seer Interactive⁵ in October 2026, found 69% of 632 B2B buyers reported inconsistencies between website information and what sellers told them.

New Map measures what engines answer about a brand and drafts the fixes. See plans and pricing.

What to do

1. Write ten buyer questions, including "How to track what Claude says about my company" and a by-name set, and run each one five times on Claude and one other engine every week.
2. Log every company named and every domain cited per answer in one sheet, grouping copies by origin, and add a column counting weekly answers that cite your own URL.
3. Compare each week with the last only after repeated runs, and correct any wrong fact at its source page, whether that is your own site or a third-party page.

Frequently asked

Sources

1. New Map measurement, How we measure, 19 October 2026
2. Otterly, AI Search in Germany Plays by Different Rules. Here Is the Data., otterly.ai, 5 October 2026
3. Otterly, AI Search in France Prefers .com Websites, otterly.ai, 12 October 2026
4. WordLift (Andrea Volpini), Why first-hit Wikidata search breaks AI agents, wordlift.io, 19 October 2026
5. Seer Interactive, Competitive Analysis for Paid Media: How To Find the Broken Stage, seerinteractive.com, 7 October 2026
6. Chen et al. (arXiv), What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization (GEO), arxiv.org, 10 October 2026