PreferiumChapter 03 of 04
What the engines answer when someone asks about your client.
Tracked prompts run on a schedule against ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews and Google AI Mode. The answer is read: which page is cited, and in which position.
4
assistants with prompt tracking
6
answer surfaces monitored
31%
citation share
58%
share of voice
The scorecard
The scorecard. Four numbers, each with its own definition.
A number without a definition is a claim nobody can verify. Two of them rest on the distinction between mentioned and cited.
Citation shareSample
31%+6
How often an AI answer uses one of the client’s own pages as a source.
The answer points at the client, and both the click and the credibility follow the link.
Share of voiceSample
58%+4
How often the client’s name is mentioned against other brands in the same answer.
It can be high while citation share is low.
Sources behind the answerSample
46
Which domains the answer actually builds on — the client’s own and others.
A domain that is fetched often and never cites the client becomes a concrete work list.
Visibility trendSample
12 weeks
The movement week by week, not a snapshot.
A category can shift without the client doing anything.
Honest measurement
A number without caveats is marketing.
Confidence intervals on everythingMention rates are shown with an uncertainty range, so one hit in two samples is never presented as 50%.
Sentiment, not just mentionsEvery mention classified as positive, neutral or negative — being talked about badly is not visibility.
Position in the answerBeing named first is worth more than being named last, so the placement is tracked, not just the presence.
Google AI Mode and AI OverviewsCitations inside Google’s conversational answers and in the summaries at the top — measured separately, never mixed with blue links.
Mentioned and cited
Mentioned is not cited. Only the last one earns the click.
Absent
The client is not in the answer at all.
“… recommends Competitor A and Competitor B.”
The model found no reason to mention the client. Often because no page on the site answers the question.
Mentioned
The name is there, but none of the client’s pages is behind it.
“… among others, Northwind Cycles.”
The model knows the client. It just sends the buyer somewhere else, because the source it trusts belongs to someone else.
CitedThe goal
One of the client’s own pages is a source in the answer.
“… according to northwindcycles.com/bike-service”
The answer points at the client. That is what earns the click, and the only one of the three states that moves anything.
Engine by engine
Engine by engine. One blended number hides who moved.
Four of them — ChatGPT, Claude, Perplexity and Gemini — are asked tracked prompts on a schedule, and we read which sources the answer was built on. Google AI Overviews and Google AI Mode are tracked separately per keyword, because they do not take a prompt the same way.
Six surfaces · different methodsMeasuring now
The six surfaces are assessed separately, each with the method that fits the surface.
ChatGPTOAI-SearchBot · GPTBotFetches at answer time and for training · measured weekly
ClaudeClaudeBot · Claude-UserSeparates training from lookups · measured weekly
PerplexityPerplexityBotAlmost always fetches at answer time · measured weekly
GeminiGoogle-ExtendedIts own control, separate from Googlebot · measured weekly
Google AI OverviewsGooglebotThe summary above the blue links · tracked per keyword
Google AI ModeGooglebotGoogle’s conversational answers · tracked separately from Overviews
Clean HTML is readable for Grok and Copilot.
The page is served as clean, semantic HTML with no JavaScript requirement. Grok and Copilot are not separate sources in the scorecard yet, so this is not a measurement result.
ProofWe ask selected questions in ChatGPT, Claude, Perplexity and Gemini — on a schedule, in your client's market.Preferium AI Edge — method
This is synthetic sampling, not insight into every conversation.Tracked answers are not the same as AI-referred traffic, and they are not every end-user conversation. Agent traffic is measured separately, at the edge — the signals are kept apart in the panel, and we never blend them into one number.
Keywords and links
AI answers are not the only thing that needs measuring.
Rankings and links are still where the demand sits — and where the models often pull their sources from. Both are tracked against named competitors, not an industry average.
Rankings and keywords28 items
Google and Bing, weeklyPositions for every tracked keyword, with daily tracking and refresh on demand. Bing is where several AI assistants retrieve from.
Volume, difficulty and intentHow many actually search each month, how hard the term will be to win, and what they want — buy, learn or find a place. With month-by-month history showing whether the term is growing or dying.
Cannibalization detectionFinds the client’s own pages competing against each other for the same term.
Keyword gap against competitorsThe terms rivals rank for and the client does not, ranked by what is actually achievable.
Position distribution and SERP featuresHow many keywords sit in the top three, top ten and beyond — and which result features appear: snippets, images, video, AI answers.
Keyword workshop and saved listsResearch any term on demand, for any site, and save the analyses in named lists you can reuse.
Traffic estimate and competitive mapEstimated traffic for the client and every tracked competitor, plotted against each other on visibility and traffic — and the keywords grouped into topics, each scored 0–100.
Site Explorer and technology detectionAnalyze any domain — competitor, prospect or acquisition target: rankings, top keywords, links, domain age, AI share and what the site is built with.
Links and authority8 items
The full link profile, weeklyTotals, referring domains and what changed since the last measurement.
Anchor text profileThe words people use when they link to the client — the quality of the profile, not just the count.
New link alertsA heads-up when a notable new site links to the client.
Lost link detectionLinks that disappeared, so you can chase them before the rankings follow.
Link opportunitiesDomains that link to competitors but not to the client yet.
Link gap against competitorsWhere rivals earn links the client misses entirely.
Referring domain profileWho links to the client, and how authoritative they actually are — not just how many there are.
Internal link opportunitiesPages on the client’s own site that should link to each other, found by meaning instead of word match.
The proof
Not our claim. Google's own data.
For every change we deploy, we pull a before-and-after window from Search Console and Analytics. If it did not work, it says so as plainly as if it did.
Per changeImpact per deployA before-and-after window with Google’s own clicks and impressions for every single change that was deployed.
The curveMarkers in the traffic graphDeployed fixes, detected content changes and rollbacks marked right in the click curve, so you see what moved what.
SourceSearch Console and Analytics dailyReal clicks, impressions, positions, click-through rates and sessions imported every day — not a number computed in-house.
TrafficEvery visit classifiedHuman, search engine, AI assistant or bot. Not just the AI share, but the whole distribution — and which assistants the visits came from.
Page and queryWho earns clicks, and who does notPerformance per page with position and click-through rate, and the actual queries that bring people to the client.
Over timeQuality snapshots per pageHow each page’s quality and performance change over weeks, not just where it stands today.
The tools
Measuring visibility is half the job. You need to see why.
Eight tools in the panel that answer why an answer turned out the way it did — instead of handing you one number and letting you guess.
SimulatorPrompt simulatorType a question and see live what each assistant answers and who is cited, with suggestions for what to fix.
InsightCrawler ViewSee any page exactly as GPTBot sees it, instead of trusting that it looks fine in your browser.
ScoreGEO scoreOne AI visibility score from 0 to 100 built on seven weighted signals, identical for every site — so scores can be compared.
ProofA/B test of citationsRun two versions of a page and measure which one the assistants prefer to cite.
EntityEntity authorityThe client’s presence in Wikipedia, Wikidata and Google’s Knowledge Graph — the sources machines trust most.
GapsAnswer opportunitiesQuestions people actually ask that the client’s site does not answer yet.
ForumsForum opportunitiesReddit and Quora threads ranking for the client’s keywords — the discussions assistants pull answers from.
MarketAds in AI answersWho pays to appear in ChatGPT answers on the client’s topics.
Local visibility
“Ranking in Maps” is not one number.
The position differs from street to street. So we measure a grid around the address instead of reporting an average nobody recognizes.
Ranking gridSeattle · 3 miles around the address
1186914742595213783261113971016
Top 34–10Outside
Sample data. Each point is an actual position in Google Maps for "bike service seattle", not an average.
Name, address and phoneWhat the client has registered, checked against what the site actually publishes — and marked up so machines can answer “who is nearby”.
Directory presenceWhether the client is listed in the leading local directories in its market, plus Facebook everywhere.
Reviews monitored weeklyTrustpilot, Google Reviews and Tripadvisor: rating, volume and movement.
QR code for Google reviewsA ready-made code that sends the customer straight to the review form in Google.
Social profiles detectedThe client’s accounts found automatically and tied to the structured data.
Want to see it on a client site?
Pick one of your client sites. We run it through the health checks, show you what the engine would fix first, and walk you through the panel your clients would see under your brand.
Agency agreementNo lock-in on your dataEvery change exportablePricing on the pricing page