Understanding snapshots

A snapshot is one row of data: one prompt, run against four engines, at one point in time. This guide explains every metric the dashboard shows and what to do with each one.

The shape of a snapshot

When you open a snapshot in the dashboard, you see five sections:

Mention — were you mentioned, yes or no, per engine
Rank — if mentioned, where you landed in the engine's list
Sentiment — how the engine talked about you
Cited URLs — what sources the engine pulled from
Share-of-voice — your mention count vs your competitors'

Each section has both a per-engine number and a cross-engine summary. The cross-engine summary is usually what you'll act on. The per-engine numbers tell you which engine to invest in if you want to move the needle on a specific platform.

Mention rate

The yes-or-no question: did the engine name your brand in its answer to this prompt?

The per-snapshot value is binary (4 yeses out of 4 engines = 100%, 1 of 4 = 25%). The useful version is the rolling mention rate over time — the percentage of snapshots in the last 7, 30, or 90 days where you were mentioned.

A rising mention rate over weeks is the single best signal that your GEO work is paying off. A flat mention rate at zero means the engines don't yet know about you for that query. A flat mention rate at 100% means the prompt is too brand-favorable and probably worth dropping.

What counts as a mention

We count a mention when the engine:

Names your brand in the answer text
Names a product you own (configured under Brand settings → Aliases)
Links to your domain in a way the parser recognizes as a recommendation

We don't count:

Mentions in disclaimers ("…such as Stripe, PayPal, or others")
Mentions where the engine is correcting itself ("not to be confused with Stripe Press")
Mentions inside a quoted user question (some engines repeat the prompt verbatim)

The match logic is conservative on purpose. False positives ruin the rolling average.

Rank

When you are mentioned, rank tells you where in the engine's list of options you appeared. Rank 1 means the engine led with you. Rank 5 means four options came first.

Rank isn't always linear. Some engines (Perplexity especially) return numbered lists, where rank is unambiguous. Others (Claude often, Gemini sometimes) return prose paragraphs that compare two or three brands inline — the parser assigns rank based on order of appearance, which is a reasonable proxy but not perfect.

For prose-style answers, treat rank as ordinal and noisy. Use the rolling average over a week, not the value of a single snapshot.

Average rank

The dashboard shows your average rank across all snapshots where you were mentioned. A 30-day average rank of 1.4 means you led the answer more often than not. An average rank of 3.8 means you're consistently in the middle of the pack.

Moving average rank from 4 to 2 is harder than moving mention rate from 20% to 40%. The engines have to actively reorder you, not just include you.

Sentiment

Three buckets:

Recommendation — the engine actively recommends your brand. Phrases like "I'd recommend Stripe for X", "Stripe is the best fit because Y".
Neutral mention — the engine names you in a list without endorsing or warning. "Options include Stripe, Square, and PayPal."
Cautionary — the engine names you with a caveat. "Stripe is good but has higher fees for certain merchant categories." Rare but it happens.

Cautionary sentiment isn't always bad — sometimes the caveat is accurate, sometimes it's stale (the engine's training data is six months old). If you see a cautionary mention with a specific factual claim attached, check whether the claim is current. If it's stale, that's actionable: the engine is repeating old information, and updating your public-facing content or pushing a fresh blog post can help correct it.

Sentiment is the noisiest metric in the snapshot. The parser uses pattern matching plus a classifier; both make mistakes on edge cases. Treat sentiment as a directional signal, not a precise measurement. If you see a sudden shift in sentiment, open the raw engine answer and read it before acting.

Cited URLs

Engines that ground their answers in retrieval (Perplexity always, ChatGPT and Gemini sometimes, Claude rarely) cite the URLs they pulled from. The snapshot lists every cited URL with:

The URL itself
Which engine cited it
Whether it's a self-citation (your own domain)
Domain authority bucket (estimated, not exact)

Why this matters more than people think: if you can identify the pages an engine routinely cites when answering your category's prompts, you have a publication target. Get mentioned on those pages — by guest posting, sponsoring, or earning a spot in a roundup — and your mention rate will follow within a few weeks.

The dashboard surfaces a "Top cited domains for this prompt" panel that aggregates citations across all snapshots. Sort it by frequency and you have your GEO content distribution list.

Share-of-voice

If you've added competitors to your brand, share-of-voice is the ratio of your mentions to total mentions across you + competitors, per prompt, per engine, summarized cross-engine.

Example: you're tracking the prompt "best CRM for small B2B sales teams" with HubSpot, Salesforce, and Pipedrive as competitors. In the last 30 snapshots:

HubSpot mentioned 28 times
Salesforce mentioned 22 times
Pipedrive mentioned 9 times
You mentioned 4 times

Share-of-voice: HubSpot 44%, Salesforce 35%, Pipedrive 14%, you 6%.

This is the most decision-useful number on the dashboard. It tells you both how visible you are and how visible the alternatives the engines are presenting to your prospects are. A growing share-of-voice means you're winning ground. A flat share-of-voice with rising mention rate means the whole category is getting more attention, but your relative position isn't changing.

Per-engine share-of-voice

Share-of-voice often varies dramatically by engine. You might have 30% share on Perplexity (which favors detailed citation-grounded answers) and 2% share on Claude (which often defaults to the largest, most well-known brand in a category).

The per-engine breakdown lets you decide where to invest. If you have low Claude share, that's a signal Claude doesn't know you yet — and the way to fix that is different from the way you fix low Perplexity share. (Perplexity is about being cited on high-authority pages; Claude is about being mentioned in widely-indexed training-eligible content.)

Reading a snapshot in three minutes

A workflow that gets to the signal fast:

Glance at mention rate. Up or down vs last week? If down, look at which engine dropped you.
Check share-of-voice. Is your relative position better or worse? If better, what changed for the competitor whose share fell? If worse, who gained?
Open the cited URLs panel. Any new domains showing up? Any of your own pages? If a new high-authority domain is being cited and you're not on it, that's a publishing target.
Read one raw engine answer. Pick the engine where your rank changed most. Read the actual paragraph. That's where you'll find the why — phrasings, comparisons, features the engine is leaning on.
Decide on one action. GEO moves slow. One concrete action per weekly review is plenty.

What snapshots can't tell you

Snapshots tell you what the engines say to a generic user. They can't tell you:

What the engines say to a logged-in user with personalization
What the engines say in a conversational follow-up (we run single-shot prompts only)
What individual humans are asking — only what your defined prompts ask
Why a result changed — the engines don't expose their reasoning

If you need attribution at the user level (which AI engine sent which converted user), that's a different tool category. Some teams pair BD GEO Tracker with referrer analysis on their site analytics to triangulate.

Snapshot retention

There's no tier-specific retention cutoff today — snapshots stay accessible for the life of your account. If we introduce a retention policy in the future, it'll be documented here and we'll give existing customers explicit notice before any data ages out.

If you want a copy of your snapshots outside our system, the cleanest path is regular CSV export. See PDFs and exports.