// Monitor · Answer Engine Insights
See yourself the way the engines see you.
Visibility, prominence, sentiment, citations and share of voice across ChatGPT, Claude, Gemini, Perplexity and Grok — computed only from answers Synthic actually received, each one stored verbatim and one click away.
5
engines, official APIs
5
buyer personas
6
prompts per keyword set
1
click from score to source
01 — The chain of custody
Every number has a receipt.
A visibility score is only worth what sits behind it. Synthic keeps the whole chain — the prompt as sent, the engine's answer as returned, the analyst verdict on that answer, and the arithmetic that turns verdicts into a score — so any figure on a chart can be opened and read.
- Prompt, model and latency recorded per answer
- Analyst verdicts: mentioned, prominence, sentiment, competitors, basis
- Score formula published, not proprietary
01
Prompt
Persona-shaped question built from your keyword.
02
Engine
Official API call to each connected engine.
03
Verbatim answer
Stored as returned — model, latency, citations.
04
Analyst verdict
mentioned · prominence · sentiment · competitors · basis
05
Score
55·mention + 30·prominence + 15·citation
02 — The record
What gets stored, field by field.
This is the row Synthic writes for each answer. Nothing is summarised before storage, and the multi-run fields exist because engines are non-deterministic: on paid plans each prompt is asked several times and the agreement between runs is stored as variance, so a lucky answer never becomes a trend.
| engine | text | which engine answered |
| model | text | the exact model id used |
| prompt | text | the question, as sent |
| response_text | text | the answer, as returned |
| mentioned | bool | did it name the brand |
| prominence | 0–100 | how early, how central |
| sentiment | enum | positive · neutral · negative · absent |
| competitors_mentioned | text[] | who else it named |
| citations | url[] | sources it cited |
| mention_basis | enum | retrieval · training |
| runs · mention_count · variance | int · int · real | multi-run agreement |
| latency_ms · capture_method · created_at | int · text · time | the evidence line |
03 — Personas
The founder and the CFO don't get the same answer.
Prompt phrasing shifts with who is asking, and so do the engines' recommendations. Run the same keyword set through five buyer lenses and compare where you win, with whom — the persona chart on the overview draws from every stored audit.
- Persona-shaped prompt generation per keyword
- Compare visibility across personas over time
- Retrieval vs training basis on every mention
01
General User
Prompts phrased as a person casually asking an AI assistant.
02
In-Market Buyer
Prompts phrased as someone actively comparing products to purchase.
03
Startup Founder
Prompts phrased as a startup founder choosing tools for a new company.
04
Developer
Prompts phrased as a technical developer evaluating tooling.
05
Enterprise Evaluator
Prompts phrased as an enterprise procurement lead.
// Where it lives
The dashboard modules that deliver it.
Every capability on this page is a working surface in the product today, not a roadmap item.
app/audits
AI Search Audits
Run live audits across every engine; every score links to its verbatim answer.
app/citations
Citations & Sources
The domains the engines cite across your audits.
app/sentiment
Sentiment
How the engines feel about you when they mention you.
app/competitors
Share of Voice
Who the engines recommend in your category, audit over audit.
// Questions
Straight answers.
How is the Visibility Score computed?
score = round(mentionRate × 55 + avgProminence/100 × 30 + citationRate × 15), over the stored answers of one audit. The formula lives in lib/scoring.ts and is shown on the site.
Are the answers real?
Yes. Each is a live API call to the engine, stored as returned. Synthic has no simulated or sample answers anywhere in the pipeline; a test enforces it.
What does 'training' vs 'retrieval' mean?
A mention backed by a live citation is grounded in retrieval — the engine looked you up. A mention with no source came from the model's training and is unverifiable; the product marks both.
The answer is being written right now.
Run a live audit and see exactly where you stand — before your competitor does.