How Do I Audit Everything ChatGPT Says About My Company?
To audit what ChatGPT says about your company, run a fixed set of roughly 25 prompts spanning five intent families — brand facts, category recommendations, comparisons, problem-solving, and reputation — in fresh logged-out or memory-disabled sessions, repeating each prompt three times. Score every answer for accuracy, sentiment, your position in any list, and which sources ChatGPT cites.
Which prompts belong in the audit?
Build five prompts per intent family, phrased the way real users type, not the way marketers search:
| Intent family | Example prompt | What you score |
|---|---|---|
| Brand facts | "What does {Company} do and how much does it cost?" | Factual accuracy |
| Category recommendation | "Best {category} tools for {persona}" | Mention + list position |
| Comparison | "{You} vs {Competitor} — which is better?" | Verdict fairness |
| Problem-solving | "How do I {job your product solves}?" | Whether you appear at all |
| Reputation | "Is {Company} legit? Any complaints?" | Sentiment + sources |
Pull phrasing from sales calls, support tickets, and Reddit threads in your niche — real buyer language surfaces different answers than keyword-style queries.
How do I keep the results clean?
ChatGPT personalizes through memory and chat history, so audit hygiene decides whether your data means anything. Use a fresh session per prompt with memory disabled (or a logged-out window), and run each prompt at least three times: answers vary between runs because sampling and retrieval are non-deterministic. Record whether the answer used ChatGPT Search (citations visible) or parametric knowledge (no citations) — the fix path differs completely between the two.
How do I score and act on the results?
Log every run in a spreadsheet with columns for date, prompt, mention (yes/no), position, sentiment (positive/neutral/negative), factual errors, and cited URLs. Two aggregate numbers fall out: mention rate (share of recommendation prompts where you appear) and accuracy rate (share of factual claims that are correct). Every factual error becomes a content task — usually a canonical page that states the correct fact plainly — and every cited third-party URL tells you which sources ChatGPT trusts in your category, which is your earned-media target list.
A manual audit is a snapshot; answers drift with every model and index update. Once the baseline exists, automated AI mention tracking re-runs the prompt set on a schedule and alerts on regressions, which is how teams catch a lost recommendation before a quarter of pipeline quietly disappears. Menra's visibility tracking automates exactly this loop across ChatGPT and other engines.
Frequently asked questions
- How many prompts do I need for a credible ChatGPT audit?
- 25 prompts across five intent families is a workable baseline for a first audit. Run each prompt 3 times in fresh sessions, which yields 75 answer samples — enough to see stable patterns without drowning in manual work.
- How often should I repeat the audit?
- Quarterly for a full manual audit, monthly if your category is competitive. Continuous automated tracking replaces the manual cycle once the stakes justify tooling, because model updates can shift answers between audits without warning.
Keep exploring
See how AI engines talk about your brand — track mentions across ChatGPT, Perplexity, Claude, Gemini and 5 more. Start with Menra