How do I know if ChatGPT or Gemini is giving accurate information about my company?
A practical audit to check whether ChatGPT or Gemini describes your company correctly, how to monitor it over time, and which tools help trace wrong claims to a source.
The short answer
You find out by asking, systematically, and comparing every answer with a written source of truth. A one-off search shows you one answer on one day. AI engines vary by model, location, wording and time, so a reliable check means running a fixed list of questions regularly, reading the responses against a fact sheet, and tracing any error back to the source it came from.
Promptwatch is the platform we would use for the monitoring part, because it records full responses and citations, has a Brand Book that defines your brand, and tracks sentiment and mentions over time. The audit itself can start today with a spreadsheet.
Step 1: write the source of truth
Before you test anything, write down what is true. Keep it short and dated.
- Legal name, location, founding date and what the company does.
- Products and plans, with current pricing or "contact us."
- Leadership names and titles.
- Certifications you can prove.
- What you do not do, because wrong "also offers X" claims are common.
If you cannot state a fact plainly, an engine will struggle too. Fix the gaps on your own site first.
Step 2: build a prompt list
Ask the questions a customer or journalist would ask, in several phrasings.
- "What is [company] and what does it do?"
- "How much does [company] cost?"
- "Who founded [company]?"
- "Is [company] good for [use case]?"
- "[Company] vs [competitor]."
- "Does [company] offer [feature]?"
- "What do people say about [company]?"
Run each in ChatGPT, Gemini and any other engine your buyers use. Capture the full text and any cited sources, not a summary.
Step 3: score each answer
For each response, mark every claim as correct, wrong, outdated, missing or unsupported. Note the position of your brand and the tone. Pricing, founders, features and competitor comparisons are where errors cluster, so give those extra attention.
Step 4: trace errors to a source
If an engine says something wrong, look at the citations. Often a wrong claim traces to an outdated third-party page, a thin directory listing, or a competitor comparison page. Fixing the model directly is not an option you control. Fixing the source is.
Promptwatch helps with this part. It stores complete responses, shows page and domain citations, tracks offsite mentions (where you are named on third-party pages without a link), and has Reddit, YouTube and LinkedIn citation reports. A Page Tracker lets you follow specific URLs, yours or a competitor's.
Step 5: monitor continuously
Accuracy drifts. A model update, a new article, or a changed pricing page can alter what engines say. A fixed panel run daily catches changes you would otherwise notice from a customer.
Promptwatch runs prompt panels daily per model, country and language. Its Brand Book holds analysis instructions that decide what counts as a mention. Sentiment is scored per response from 1 to 100 with a weekly AI summary, and Visibility Score, Share of Voice and Position in Answer give the trend. Brand aliases help it recognize name variants.
On models: Promptwatch monitors ChatGPT (Search) and Google Gemini through the live interface, plus Google AI Overviews, AI Mode, Copilot, Perplexity and others. It tracks Claude, DeepSeek, Mistral, Grok and some other models via API, and says so.
Tool comparison
| Tool | Useful for accuracy checks | Gap | Starting price |
|---|---|---|---|
| Promptwatch | Full responses, citations, Brand Book, sentiment, offsite mentions | Crawler logs start on Professional | Free tier; paid from $95/mo |
| AthenaHQ | GEO score with sentiment analysis, action center | Credit-based billing | Free tier; Starter $295/mo |
| Scrunch AI | Sentiment analysis, site audits | Weekly refresh per our entry | Core $250/mo |
| Profound | Answer-level analytics, citation share | Enterprise-gated; about $40k a year | Trial; Enterprise custom |
| GEO Monitor | Full answer text with brand highlighted | Very small vendor; no Perplexity | From €69/mo |
Prices come from the catalog entries and may have changed.
Who should pick which
- You want ongoing monitoring with citations and a brand definition: Promptwatch.
- You want a daily action queue to fix issues: AthenaHQ.
- You are a French-speaking small business wanting full answer text: GEO Monitor, with its support caveat.
- You only need a one-time check: a spreadsheet and thirty minutes will do.
What a tool cannot do
It cannot make an engine correct itself. It shows you what was said and which sources fed it. Fixing the claim means fixing the source, then waiting for engines to refresh. Some errors will not change at all, and some engines will keep citing an old page for a while. For anything touching legal, medical or financial claims about your company, get professional advice and document corrections carefully.