Platform

Measure what AI tells your buyers, then change it

Six connected surfaces: what the assistants answer, which questions are worth winning, who wins the ones you lose, which sources decide it, what to do next, and how to report it. Every number traces back to a stored response.

Engines measured via provider APIs
6

Engines measured via provider APIs

Buyer-intent categories
13

Buyer-intent categories

Funnel stages
5

Funnel stages

Action stages, new to verified
7

Action stages, new to verified

AI visibility

Know which buyer questions mention you, and which ignore you

Share of answer, visibility index, citations and average position — split by engine and by market, tracked over time rather than screenshotted once.

  • Share of answer over time, plotted against your category average
  • Visibility by engine — mention count and share for every engine on your plan
  • Visibility by market — prompts, runs, mention rate and average score per market
  • The prompts you appear in most, and the open actions attached to them
How these numbers are derived

How the visibility index is built

Deterministic, and the same every time

  • Mention rate30%
  • Position score20%
  • Citation ownership20%
  • Sentiment15%
  • Share of AI voice15%

Computed by our own code from your stored runs. No model is ever asked to pick a score, so the same runs always produce the same number and you can check the arithmetic yourself.

What makes a prompt worth winning

Prompt value score, out of 100

  • Buyer intent18%
  • Conversion potential12%
  • Revenue relevance12%
  • Competitor intensity12%
  • Search demand12%
  • Funnel weight12%
  • Product relevance8%
  • ICP fit8%
  • Region importance3%
  • Strategic importance3%

Weights sum to 100. Signals you have not connected fall back to neutral rather than being invented, so an unconnected workspace is scored conservatively instead of optimistically.

Buyer prompts

Spend your effort on the questions that carry pipeline

Not every question is worth winning. Prompts are scored on buyer value, so a pricing comparison and a definition lookup never compete for the same attention.

  • A 13-category intent taxonomy, from pricing and buying through comparison, alternatives and competitor replacement down to informational
  • Five funnel stages, from awareness to post-purchase
  • Value tiers — high at 70 and above, medium at 45 and above — so the portfolio sorts itself
  • Portfolio views for the cases that matter: high-value lost, win, and competitors own
  • A prompt-by-engine matrix showing your position on each engine and its movement against last week
  • Write your own prompts, import them, or start from suggestions generated for your brand and category when you onboard
Outcomes tracked per prompt
  • Win
  • Lost
  • Competitor owns
  • Neutral
Competitor intelligence

See who replaces you in an answer, and why

Displacement analysis turns a competitor appeared and you did not into the specific claims, sources and content patterns that put them there.

  • A share-of-answer leaderboard with your rank in the field, and your own row marked
  • Head-to-head across AI visibility index, citation share and prompt coverage — with the gap and the winner on each row
  • Why they win: the verbatim claims the assistant made about them, quoted from the response
  • Source evidence per finding — domain, source type, whether it supported the claim, citation count and the engine that cited it
  • Source overlap: which domains cite both of you, and which cite only them
  • A recommended action attached to every finding, with its priority

Best tools for enterprise onboarding

Absent
High confidence · 48 runs · 3 enginesPriority 82
Recommended instead
Two competitors, named in 44 of 48 runs
Why they win
Both are cited from a comparison page on a review domain that never mentions you.
Source overlap
12% shared domains
Illustrative example. Findings are generated from your own prompt runs.
Citations

Know which sources are shaping the answer

An assistant that cites tells you where the answer came from. TrackGeo records every cited URL, classifies the domain, and shows whether it is working for you or for a rival.

Every cited source, kept

Source URL, engine, position, capture time and type — recorded from the run itself, not reconstructed later.

Source types, classified

Editorial, review, community and owned. Knowing whether to earn coverage or write it yourself depends on this split.

The source mix behind your category

Trusted, community, social, owned and other — with how many are yours and how many belong to rivals.

Top cited domains

Which domains assistants reach for when answering your category, ranked by how often they appear.

Citations over time

Total citations, citation share, cited domains and citation rate — as a trend, so you can see a source arrive or disappear.

Opportunity gaps

Domains cited for your rivals but never for you. This is opt-in and never polls in the background.

Engines differ in what they expose. Some return sources alongside the answer, which is what makes source tracking possible on those surfaces. Where an engine returns no sources, we record what it did return rather than inferring a citation.

See engine coverage
Actions

Turn every gap into work someone owns

A score that drops does not tell you what to fix. Every finding becomes a prioritised action with evidence, an owner and a due date.

  • A priority score out of 100, so the backlog arrives already sorted
  • Seven stages from new to impact verified, as a list or a Kanban board
  • Filter by type, priority, difficulty, competitor, prompt type, owner, and whether TrackGeo raised it or you did
  • Evidence, why it matters, and suggested fix steps on every action
  • A copyable JSON-LD snippet for schema work, ready to paste
  • Owners, due dates and notes — it is a backlog, not a list of observations
See how actions reach your team

How priority is calculated

Action priority score, out of 100

  • Business impact35%
  • Visibility opportunity25%
  • Competitive gap20%
  • Ease of implementation10%
  • Confidence10%

Confidence is a term in the formula, which means a finding from a thin sample cannot outrank a well-evidenced one.

The lifecycle

  1. New
  2. Accepted
  3. In progress
  4. Implemented
  5. Monitoring
  6. Re-scan needed
  7. Impact verified

Impact verified is earned, not asserted. Mark an action implemented and TrackGeo re-scans the related prompt after the follow-up window, then records the before and after. Until that happens the result stays empty.

Reports

Give your team and your clients something they will read

Scheduled briefs, exports and white-label reporting — generated from the same runs and the same evidence as everything else.

Reports

Five report types — executive summary, citations, competitors, prompts and runs. Generate any of them on demand over a 7, 30 or 90-day window, and export as CSV. Automated delivery on a schedule is not available yet.

  • On demand
  • Weekly
  • Monthly

The AI Pipeline Gap Sprint

The high-value prompts you are losing, who wins them, why, and the sample behind each — followed by a prioritised backlog and a 30-day plan. It carries its own confidence and methodology section, so it can be handed to someone who was not in the room.

  • CSV
  • Sources CSV
  • HTML
  • PDF

Agencies and white label

Link client workspaces and see visibility, movement and credit usage per client in one table. White-label reports carry your report title, your logo and your accent colour.

  • Report title
  • Logo
  • Accent colour

Reports are a paid feature. See which plans include them.

Evidence

Every finding shows the sample it came from

Assistants are non-deterministic. One answer is an anecdote, and a product that presents it as a fact is selling you a screenshot.

Repeated runs

Your prompts run on a schedule, daily or weekly, so a pattern can separate itself from a coincidence.

Confidence gated by sample

Confidence blends how consistently a finding repeats with how much evidence backs it. Fewer than two runs is always low confidence — consistency is undefined without repetition.

Market and language are separate

The same question gets a different answer in a different market. Those are separate measurements, never averaged together.

Read how prompts are selected, how confidence is derived, and what we explicitly do not guarantee.

See our methodology
Also in the platform

The rest of the surface area

Depth where it changes a decision, and the plumbing to keep it running without you.

Sentiment and brand health

How assistants characterise you on a −100 to +100 scale, the negative drivers behind it, and a monitor for negative or outdated claims.

Site intelligence and audits

A GEO score for your site, pages scanned, and open issues by priority — each with a description and a recommendation.

AI traffic

Assistant referrals separated into observed, estimated and inferred. What we saw is never blended with what we modelled.

Daily brief and alerts

What changed, why, and what needs attention — quiet when nothing meaningful moved.

Questions

What people ask before they trust a number

Is the visibility score generated by an AI model?

No. Every score in TrackGeo is computed by our own deterministic code from your stored runs. Models are used to run your prompts, classify sentiment, extract entities and write prose explanations — they are never asked to pick a score. That means the same runs always produce the same number, and you can check the arithmetic against the published weights.

How do you decide which prompts are worth tracking?

Each prompt gets a buyer-value score out of 100, blending buyer intent, conversion potential, revenue relevance, competitor intensity, search demand, funnel stage, product relevance, ICP fit, region and strategic importance. Prompts scoring 70 and above are tiered high, 45 and above medium. A pricing comparison and a definition lookup are not worth the same effort, so they are not treated alike.

What does displacement analysis actually show me?

It turns the fact that a competitor appeared and you did not into the reason it happened. Each finding names the prompt, the competitors recommended instead, the verbatim claims the assistant made about them, the sources behind those claims, how much your source set overlaps with theirs, and a recommended action — with the sample size and confidence attached.

Can I see the answer behind a finding?

Yes. Every finding opens onto the response the assistant actually gave, the engine that produced it, when it was captured, the sources it cited, and what changed since the previous run. Evidence depth varies by plan, but no finding asks you to take our word for it.

Do you show whether an action worked?

Only once a re-scan has verified it. When you mark an action implemented, TrackGeo re-scans the related prompt after the follow-up window and records the before and after. Until that happens the result stays empty rather than being filled with an estimate.

Find the buyer questions your competitors are winning

Start with a free audit of your site, or talk to us about monitoring your brand, market and competitors continuously.