Most teams asking “how do we show up in ChatGPT” have never looked at what it currently says about them. That’s the first thing to fix, and it costs nothing but an afternoon.
Build a prompt set that reflects real buying
The instinct is to type your brand name and see what comes back. That’s the least useful prompt you can run — the model will find you, and you’ll conclude everything is fine.
Buyers don’t start with your name. They start with their problem. A useful prompt set covers four categories:
- Category discovery: “best [category] tools for [segment]” — where shortlists are formed.
- Comparison: “[competitor] vs alternatives”, “who competes with [competitor]”.
- Objection: “is [your brand] worth the money”, “problems with [category] tools”.
- Branded: “what does [your brand] do”, “[your brand] pricing”.
Sixty prompts is a reasonable baseline. Fewer and single-run randomness dominates your results.
Run each prompt more than once
These systems are non-deterministic. The same question can produce different answers minutes apart. Running once and treating the output as fact is the most common error we see in DIY audits.
Run each prompt at least three times, in a fresh session with no history, and record all outputs. What you care about is the rate at which you appear, not whether you appeared once.
Score what you get back
For each response, record four things: were you mentioned at all; were you cited with a link; was the sentiment positive, neutral or negative; and which competitors appeared instead. That’s enough to compute a mention rate and a share-of-answer figure per engine.
Trace the errors to their source
This is the part most audits skip, and it’s where the value is. When an engine says something wrong about you, find out where it read that. Usually it’s one of: an outdated page you forgot to retire, a directory listing with stale data, a review site summary, or a competitor’s comparison page that went unchallenged.
Fixing the source fixes every prompt downstream of it. Arguing with the model does nothing.
What “good” looks like
There’s no universal benchmark — it depends heavily on category maturity. As a rough orientation: appearing in under 20% of your category prompts means you’re effectively invisible; 40–60% is a normal position for an established brand doing no deliberate GEO work; consistently above 70% usually indicates someone has been working on it.