How to check what ChatGPT, Claude & Gemini say about your company

Your buyers already ask AI assistants about you — for shortlists, comparisons and "is this company legit?" checks. Here's how to audit those answers yourself, and the mistakes that make a casual check misleading.

1. Ask buyer-shaped questions, not vanity questions

"Tell me about [company]" is the question a founder asks; it's rarely the question a buyer asks. Buyers ask in their own context:

• "What are the best [category] tools for [use case]?" — are you on the shortlist at all?
• "Compare [you] vs [competitor] for a [size] team."
• "Is [company] reliable? Any problems I should know about?"
• "Which [category] vendor would you pick for [requirement], and why?"

That last shape matters most: asking the model who it would pick and why surfaces quiet deferrals — cases where it recommends a competitor using strengths you actually have.

2. Run each question more than once

Model answers vary between runs. One response that mentions you proves little; the question is what the model says usually. Ask each question several times in fresh conversations (so earlier turns don't steer the answer) and treat anything that appears in most runs as the signal. We use three runs per question per model as the floor.

3. Cover more than one assistant

ChatGPT, Claude and Gemini are trained and grounded differently and routinely disagree about the same company — one may describe you accurately while another has you confused with a similarly-named firm. A single-assistant check is a third of an audit.

4. Grade against your facts, not your feelings

Write down your verifiable facts first — what you do, certifications, locations, products, real differentiators. Then grade every claim in every answer against that list: wrong (fabricated or outdated), missing (a differentiator the model never mentions), or right. Severity beats volume: a fabricated compliance problem outranks a wrong founding year every time.

5. Fix the fetchable layer

You can't edit a model's training data, but you control what its crawlers fetch — and that's what updates next crawl. In order of effort: make sure robots.txt isn't blocking AI crawlers (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended); publish an llms.txt fact sheet at your site root (free generator here); add structured data and FAQ content that states your facts unambiguously. The background on why these work is in our GEO explainer.

The honest caveat

Done properly — say 10 buyer questions × 3 models × 3 runs — that's 90 conversations to read and grade, and you'll want to repeat the whole exercise after every fix. The method is genuinely this simple; the discipline is the hard part. That's the gap a ReputationIQ scan closes: the same battery, run and graded automatically against your verified facts, ranked by severity, with the fixes drafted for you.