AI Discovery Transcripts
The AI Discovery Engine measures whether large language models — ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews — recommend your app when someone asks them for one. The headline scores tell you how you’re doing. The transcript tells you why: the exact prompts each model was asked, and the raw answer each one gave back. This page shows how to open that detail and how to read it.
Before you start
- Open a completed audit and go to its AI Discoverability section.
- Full per-model detail is plan-gated. On the Free and Basic plans you can open the transcript for the first model; opening the others prompts an upgrade to Pro. See Plans and entitlements for what your plan includes.
Open a transcript
1. Go to the Per-LLM Breakdown
In the AI Discoverability section, find the Per-LLM Breakdown — one card per model, each showing its presence, coverage, and dominant sentiment.
2. Open the detail
Click Details on a model’s card (or a row in the Cross-LLM Panel). A dialog opens with that model’s prompt battery and the stored response for each prompt.
3. Read the threads
Each entry in the dialog is one prompt paired with the model’s raw answer:
- The prompt — the exact question the model was asked (for example, a category or use-case query a real user might type).
- The raw response — the model’s full answer, verbatim. Where your app is mentioned, the mention is highlighted inline.
- Mentioned / No mention — a badge on each response saying whether your app came up at all.
- Detected mention — when your app is mentioned, the surrounding snippet is pulled out so you can read the context without scanning the whole answer.
- Sentiment and position — for a mention, whether it read positive, negative, or neutral, and roughly where in the answer it landed.
- Run number — some prompts are asked more than once to check consistency; each run is labeled so you can see whether the model mentions you every time or only sometimes.
A summary strip at the top counts the prompts, the mentions, and the responses with no mention, so you get the shape of the model’s behavior before reading individual threads.
How to read it
The transcript is the evidence behind the score. Use it to answer questions the score alone can’t:
- Why is my presence low for one model? Read its threads. If it never mentions you across relevant prompts, that’s a coverage gap. If it mentions you inconsistently across runs, your presence is fragile rather than absent.
- Is a mention actually good? A mention with negative or hedged sentiment, or one buried at the end of a long list, is worth less than an early, confident recommendation. The sentiment and position tell you which kind you got.
- What language do the models use? The prompts show how models frame the category your app competes in, and the responses show which competitors they name — useful signal for your own positioning.
Troubleshooting
The Details button prompts me to upgrade
Full per-model transcripts are plan-gated. The first model’s detail is available on lower plans; the rest require a plan that includes full AI Discoverability detail. See Plans and entitlements.
The dialog says the transcript is unavailable
That audit doesn’t have a saved per-prompt payload for this model, so the raw prompts and responses can’t be rendered. This happens on older audits. Run a new audit and open its transcript instead.