Skip to content
Get started

AI search visibility

Find out what AI tells buyers about you.

SeenInSearch puts the questions your buyers actually ask to ChatGPT, Google AI Overviews and Perplexity — then scores how often you appear, where you rank, and who gets recommended instead.

Free · No account needed · Every weight published

ChatGPTacme.com

A buyer asks

Best project management tools for a small design team?

The answer names

For a small design team, Northwind is usually the first recommendation for its visual planning, with Acme a solid alternative when you need tighter approval workflows.

Your rank
#2
Cited
g2.comreddit.com
Verdict
Named second
1/3

Answer engines

  • ChatGPT
  • Google AI Overviews
  • Perplexity
  • Gemini
  • Claude
  • Google AI Mode

Gemini, Claude and Google AI Mode are not queried yet — nothing they say counts toward your score.

The blind spot

You can read every review, every ranking and every dashboard you own — and still have no idea what an assistant says about you the moment a buyer asks.

  • No ranking to check
  • No referrer
  • No signal at all

Why it matters

Search stopped being a list of links.

Answer engines do not return ten options for the buyer to weigh. They return a recommendation. Being in it is the whole game.

  1. 01

    A buyer asks an assistant

    Instead of opening ten tabs, they ask one question in plain language and wait for an answer.

  2. 02

    The assistant answers with a shortlist

    Three or four names, in an order, with sources attached. Most people never scroll past it.

  3. 03

    That shortlist is the consideration set

    If you are not named, you are not evaluated. There is no second page to rank on any more.

  4. 04

    And it decides who gets the deal

    Being absent from the answer removes you before a conversation ever starts — silently.

The scorecard

The whole picture on one page.

Every scan produces the same report: one number, the five sub-scores behind it, the competitors named instead of you, and the engines it was all measured across. Switch engines below — the breakdown and the headline recompute from the same weights /methodology publishes.

acme.com

AI Visibility Score

61

Visible

You show up in AI answers, but not always first.

  • Presence 35% weight71
  • Prominence 20% weight58
  • Share of voice 20% weight44
  • Citations 15% weight55
  • Sentiment 10% weight78

Share of voice

  • You4
  • Northwind3
  • Contoso2
  • Globex0

Scored by engine

Engine weights are renormalised over the engines that actually answered — pick one to see its own breakdown.

Citations

Every answer leans on something.

One buyer question fans out across the engines, and each answer cites its sources. We track which domains those are — and whether yours is ever one of them.

Citations behind an answerA network. One large node is a single AI answer; four smaller nodes hang below it on curved links — the domains that answer cited. Each node's size and the weight of its link show how much of the answer's citation list pointed at that domain, and the share is printed beside each node: g2.com 36 percent, reddit.com 29 percent, vendor docs 21 percent, your own site 14 percent. Your own site is the highlighted node: whether an answer cites you at all, rather than only third parties, is what the citations sub-score measures. The domains and the counts are illustrative.g2.com36%reddit.com29%your site14%vendor docs21%ONE ANSWERTHICKER — CITED MOREILLUSTRATIVE SOURCES
We read what was attributed
Citation domains come straight out of the engines’ own responses. Nothing is inferred from a link we went and found, and no cited URL is fetched or verified — we report what the answer said it leaned on.
Being cited is not the same as being named
An answer can recommend you without linking you, or link a review site that happens to mention you. Those are different failures with different fixes, so they are measured as separate sub-scores.
Third parties are the usual answer
Most answers lean on review sites, forums and documentation rather than the vendor's own pages. Which of those you can realistically influence is what the recommendations further down this page are for.

The full fan-out

The same mechanic at the scale a scan runs it: one question, every engine, and the sources behind each answer. Hover any node to isolate its path.

BUYER QUESTIONANSWER ENGINESSOURCES CITEDBest project managementtool for a design team?ChatGPTGoogle AI OverviewsPerplexityg2.comreddit.comcapterra.comacme.comYOU

Illustrative structure, not a live scan. Citation domains come straight from the engines' own responses — we read what they attributed, we do not infer it.

What you get

Six measurements, no dashboard archaeology.

Each scan returns the same set of signals, so the number means the same thing every time you run it.

  • AI visibility analysis

    How often you appear across the buyer questions we ask, and how high you rank when you do.

  • Share of voice

    Every competitor the answer named alongside you, tallied question by question.

  • Citation tracking

    Which sources the answers leaned on, and whether your own domain was one of them.

  • Sentiment read

    Whether a mention was positive, neutral or negative — not merely that it happened.

  • Per-engine breakdown

    The same five scores, engine by engine, so you can see exactly where you go missing.

  • Deterministic scoring

    Scores are computed in code, not written by a language model. The same run gives the same number.

The question set

The questions are yours, not a template.

A visibility score is only worth the questions behind it. Yours are generated from what your homepage says you sell, then put to every engine — 6 on a full report, 3 on the free teaser.

The buyer question set6 buyer questions splayed like a fanned hand of cards: one thin blade per question, all the same size, all pivoting from a point below the frame. The 3 blades at the leading edge of the fan, nearest the viewer, are outlined and washed in green — those are the questions a free teaser scan also runs. A bracket across the base of the fan spans all 6 and turns green across the teaser's share of the spread. A full scan runs every blade.3/66 QUESTIONS3 IN THE FREE TEASER
Written from your own homepage
We fetch your site first and infer what you sell from it. The questions are about your category, not a generic template — and if your site can't be reached, the scan says so rather than guessing quietly.
High intent, not your brand name
Not “what is Acme”. The questions are the ones a buyer types when they are choosing between options and have not decided yet — which is the moment where being named actually changes anything.
Every question, to every engine
All 6 go to each of the 3 answer engines — 18 answers in a full scan. One engine liking you is not a result.

Answer engines

Where the questions actually go.

Every question goes to the 3 engines we query. The other 3 are drawn, not queried — so they cannot move your number.

The answer engines a scan queriesOne scan at the centre with the answer engines in orbit around it. ChatGPT, Google AIO and Perplexity are filled discs on the solid inner orbit: those have adapters and are queried on every full scan. Gemini, Claude and Google AI Mode are smaller hollow rings on a dashed, ghosted orbit further out — they carry weights in the scoring table, but nothing queries them today, so they are roadmap rather than coverage.60°ONE SCANChatGPTGoogle AIOPerplexityGeminiClaudeGoogle AI ModeSOLID — QUERIEDDASHED — ROADMAP
Solid means we query it
ChatGPT, Google AI Overviews and Perplexity each have a validated adapter. Every full scan puts all 6 questions to all 3 of them, in the same run.
Dashed means roadmap, not coverage
Gemini, Claude and Google AI Mode are drawn as outlines because no scan queries them: the adapters exist, but none has been run against the live API, so nothing they would say counts toward your score. A diagram is a worse place for a false capability claim than a sentence is.
Scored engine by engine
The same five sub-scores are computed per engine, so you can see exactly which one you go missing on. The report also records which engines answered and which failed, rather than quietly dropping them.

How it works

Four steps, about a minute.

The answer phase is capped at 85 seconds. Nothing about the run is hidden — the report records which engines answered and which failed.

  1. 1

    Enter your domain

    No account and no card. We read your homepage first, so the questions are about what you actually sell.

  2. 2

    We ask what your buyers ask

    6 buyer questions, put to ChatGPT, Google AI Overviews and Perplexity — 18 answers in all.

  3. 3

    The score is computed in code

    Five sub-scores with fixed, published weights. A language model reads the answers; it never sets the number.

  4. 4

    You get the report

    Where you ranked, who was named instead, which sources were cited, and the fixes worth doing first.

Inside a run

Five stages, in the order the worker runs them.

Above is what you do. This is what happens after you hit scan — ask, collect, parse, score, publish — and the order is not editorial, it is the order the code executes.

A scan from question to published reportOne continuous conduit carrying a single run of work from left to right. Its bore steps down at each of five machined bulkheads, in the order the worker executes them: Ask, Collect, Parse, Score, Publish. It enters at full bore — every question, on every engine — and the last section, drawn in green, is the single published report leaving the outlet.0102030405ASKCOLLECTPARSESCOREPUBLISH
Ask and collect are time-boxed
The answer phase is capped at 85 seconds. An engine that fails inside that window is recorded as failed and excluded — the report never quietly presents four answers as if they were six.
Parsing and scoring are different jobs
A language model reads the answers and extracts what was said. The score is then computed in code from fixed, published weights — the model never sets the number, which is why the same run gives the same result.
Publish is the last stage, not an email
The run ends at a report on its own URL, which is why the final volume in the figure stands taller than the four before it. Nothing is held back for a follow-up.

Example report

Know exactly what lands before you run one.

This is the anatomy of a full report, filled in with an illustrative scan. Nothing here is a mock-up of a feature that does not exist.

Overview

acme.com scored 61 out of 100 — Visible. Named in four of six buyer questions, usually second behind Northwind.

Visibility score
61Visible

You show up in AI answers, but not always first.

Strengths
  • Named in 4 of 6 buyer questions
  • Tone is consistently positive when you do appear
Weaknesses
  • Ranked second or lower in most answers
  • Your own domain is cited less often than review sites
Competitors
  • Northwind3 of 6
  • Contoso2 of 6
  • Globexnot seen
Verbatim answer

What are the best project management tools for a small design team?

For a small design team, Northwind is usually the first recommendation for its visual planning, with Acme a solid alternative when you need tighter approval workflows. Contoso is worth a look if budget is the main constraint.

Ranked 2 of 3 named brands

Citation sources
  • g2.comReview roundup
  • reddit.comCommunity thread
  • capterra.comCategory listing
  • acme.comYour own site
Recommendations
  • Publish a comparison page against Northwindhigh impact · medium effort

    You are named second in most head-to-head answers. A direct comparison page gives the engines something to cite when the question is framed as a choice.

  • Get listed in the category roundups AI already citeshigh impact · low effort

    Three of the four sources behind these answers are review directories. Your absence from two of them is why citations trails presence.

  • Add structured data to your product pagesmedium impact · low effort

    Clear product markup makes it easier for answer engines to attribute a claim to your own domain rather than to a third-party listing.

Share of voice

Your number only means something next to theirs.

A visibility score on its own is a number without a scale. What makes it actionable is who else the answer named, how often, and how far ahead of you they were.

Your visibility ranked against the competitors an answer namedIllustrative figures from the sample report, not a live scan. 4 brands ranked by visibility score on a shared 0 to 100 baseline, highest first: Northwind 74, You 61, Contoso 52, Globex 38. The highlighted row is yours — rank 2 of 4, 13 points behind Northwind.74615238NorthwindYouContosoGlobex01020304050100+13ILLUSTRATIVE — SAMPLE REPORT
The competitors come from the answers
You do not supply a list. Whoever an engine names alongside you — or instead of you — is who you get measured against, tallied question by question across the run.
One baseline, so the gap is readable
Every row is measured from the same zero, with a line carried down the set at your own score. Being thirteen points behind is a different problem from being two, and the shape should say which one you have.
It has to be able to show you losing
The highlighted row in the figure is not the longest one. Share of voice carries 20% of the overall score precisely because it is the part you cannot fix by talking about yourself more.

Figures shown are illustrative and match the example report above — not a live scan.

What to fix first

Every fix arrives with a price on it.

A list of everything wrong is a report you close. Each finding carries an impact and an effort, so the cheap high-return work is visible at a glance instead of buried three bullets down.

Findings placed by impact against effortA scatter plot of recommendations, on the only two fields a finding is stored with. Effort runs left to right in 3 steps — low, medium, high — and impact runs bottom to top through the same 3: low, medium, high. Those words are the whole vocabulary; neither field takes any other value, so there are 9 places a mark can be and no others. 3 findings from the example report are drawn as circles at those positions. A dashed cross divides the plane where the vocabulary divides it, and the shaded top-left quadrant is high impact for low effort — the work worth doing first. The marks inside it, 1 of 3, are filled discs in green; the rest are hollow rings, because they either cost more or return less.DO FIRSTIMPACTEFFORTLOWMEDIUMHIGHHIGHMEDIUMLOWILLUSTRATIVE — SAMPLE REPORT1 OF 3
Two fields, not a priority theatre
Every finding is stored with an impact and an effort, and that is the whole ranking model. No weighted urgency index, no colour-coded tiers over the top of a guess.
Cheap and high-impact is a position, not an opinion
A report returns five fixes. Reading them against both axes is what separates the one worth doing this afternoon from the one worth a quarter — the highlighted marker is the most impact for the least work.
Written from your run, not a checklist
The fixes are drafted from what the scan actually saw — the questions you were missing from, the sources that cited someone else, the engines that never named you. A generic AEO checklist would not need your results to produce.

Methodology

You can check our arithmetic.

The score is not a black box and not a vibe. Five signals, fixed weights, computed in code — so the same scan always produces the same number, and you can see precisely which part is dragging you down.

Weighted contribution to the overall score100%
How the five sub-scores combineA single gauge ring divided into five arc segments, one for each sub-score. Every segment's share of the ring is that sub-score's weight in the published scoring table: Presence 35%, Prominence 20%, Share of voice 20%, Citations 15%, Sentiment 10%. Presence is the largest segment and the only one drawn in green; Sentiment is the smallest. A fine scale marks the running total at each break — 35, 55, 75, 90, 100 — so the ring closes at one hundred. At the centre, the number the five weighted parts resolve to: 61, a sample reading in the Visible band.35557590100Presence35%Prominence20%Share of voice20%Citations15%Sentiment10%Sample61Visible
  • Presence35%

    Do you show up at all across the buyer questions we ask.

  • Prominence20%

    When you are mentioned, how high you rank among the names given.

  • Share of voice20%

    Your mentions weighed against every competitor the answer also named.

  • Citations15%

    How often the answer links back to your own domain rather than a third party.

  • Sentiment10%

    How positively you are described when you do get mentioned.

Trust

Reasons to believe the number.

A visibility score is only worth as much as the process behind it. Here is what that process actually guarantees.

  • Scores computed in code

    A language model reads the answers. It never decides the number — that happens in a scoring function with fixed weights.

  • Weights published in full

    The methodology page imports the same constants the scorer runs on, so the published numbers cannot drift from the real ones.

  • You control publication

    Reports are public by default at a URL derived from your domain. If you own the scan, you can unpublish it in one click.

  • No third-party analytics

    No trackers, no ad pixels, no session recording. Product events are written to our own database and nowhere else.

  • No account for the free scan

    We ask for an email only when you want the full report — and then only to send it to you.

  • Failed engines are disclosed

    Every report records which engines answered and which did not, so a thin scan can never pass itself off as a complete one.

FAQ

Questions worth asking first.

How is the score calculated?
Five sub-scores — presence, prominence, share of voice, citations and sentiment — are each measured 0–100 and combined using fixed weights. The arithmetic runs in code rather than in a language model, so the same inputs always produce the same score. Every weight is published on the methodology page.
Which AI platforms do you analyse?
The full report covers ChatGPT, Google AI Overviews and Perplexity. The free teaser scan uses ChatGPT only. Gemini, Claude and Google AI Mode are on the roadmap — they are not measured today, and we do not count them in your score.
How accurate is the report?
It is a sample, not a census. Answer engines are non-deterministic and will phrase things differently from one run to the next, so read the score as an observation of how you are currently described rather than a guaranteed measurement. Scans currently run in US English.
Can I compare myself against competitors?
Yes. Every brand named alongside you is tallied across the buyer questions, and your share of voice is one of the five sub-scores. Competitors that were expected but never actually appeared are shown at zero rather than quietly dropped.
How often is the data updated?
Each scan is a fresh run against the live engines — there is no cached corpus behind the score. Free teaser scans for the same domain are reused for 24 hours so a repeat visit does not trigger a second billable run.
Do you store my data?
Yes, and specifically: the scan, the verbatim answers the engines returned, and the IP address that started it. We do not use third-party analytics — product events go to our own database. Reports are public by default at a URL derived from your domain, and the owner can unpublish at any time.
Do I need an account?
Not for the free scan — enter a domain and you get a result. A work email is required only for the full report, so we have somewhere to send it.
How long does a scan take?
The answer phase is capped at 85 seconds and most scans finish well inside that. You will see live per-engine progress while it runs, rather than an indefinite spinner.

Find out where you stand.

One domain, one scan, no account. You will know within a minute whether answer engines are recommending you — or recommending someone else.