DavidGadyan

AI Search KPIs: 12 Ways to Measure AEO and GEO

AI search KPIs: twelve ways to measure AEO and GEO, from visibility rate and share of voice to self-reported attribution, and what to do when the numbers move.

  1. Metrics

    Visibility rate

    Are you in the answer?

    When it helps
    Nobody knows whether ChatGPT, Perplexity or Google AI Overviews mention the brand at all, or only at the decision stage.
    What to do
    Track the share of relevant AI answers that mention you, split by topic, funnel stage and customer segment, and read it weekly per engine; it is the foundational metric everything else builds on.
    Why it works
    Without a visibility rate every other number is guesswork; split by category it shows where you are visible and where competitors are winning, not just one aggregate score.
    Example
    You appear in 35 of 100 tracked answers: visibility 35 %. Split: 60 % in "how to" prompts, 10 % in "best tools for" prompts, which is where buyers compare.

  2. Metrics

    Average position

    First or fifth?

    When it helps
    The brand is mentioned often but always after four competitors, and the visibility score flatters the real competitive position.
    What to do
    Record where you are named in each answer, aggregate weekly across many prompts, and treat the first two sentences of an answer as the new page one.
    Why it works
    Brands named first or second get disproportionately more attention than those listed fifth or tenth; day-to-day results swing because models are non-deterministic, so only weekly averages show the trend.
    Example
    "Which agencies build AI agents in Barcelona?": named first in 3 answers, third in 5, not at all in 2. Average position 2.3 this week, 3.1 last month: gaining ground.

  3. Metrics

    Sentiment and its sources

    What AI says about you

    When it helps
    You are visible but not converting: the AI mentions the brand with caveats ("expensive", "hard to set up") that come from somewhere.
    What to do
    Score the tone of AI mentions by funnel stage, track it monthly, and attribute it to the domains the models cite; evaluation-stage prompts ("is X reliable?", "is X worth the price?") come first.
    Why it works
    Sentiment is a leading indicator of conversion, and unlike training data the sources behind it can often be fixed quickly.
    Example
    In the post's case, 60 % of a brand's negative AI mentions traced to one review aggregator with single-submission accounts; one legal letter resolved it in two weeks and sentiment moved within 30 days.
    Reported figure
    Sentiment moved within 30 days of fixing one source (KIME case) (estimate)

  4. Metrics

    Share of voice

    You vs named rivals

    When it helps
    A 40 % visibility rate looks fine until you learn the nearest competitor sits at 70 %, or looks weak when nobody else exceeds 25 %.
    What to do
    Benchmark your mentions against the three to five brands you really compete with, on the same prompts, per category and per engine, every week.
    Why it works
    Competitive context is how you read your own data; share of voice is only meaningful against the right rivals.
    Example
    You get 40 of 100 brand mentions in your category: share of voice 40 %. Per engine: 55 % on Perplexity, 20 % on Gemini, so Gemini's preferred sources are the gap.

  5. Metrics

    Citations

    Which pages shape the answer

    When it helps
    The AI describes your product from a two-year-old review instead of your own pricing page.
    What to do
    List the web pages each engine relies on or links to when it answers about you (your own pages, review sites, comparison articles, media), then check whether they are accurate and current.
    Why it works
    Drops in share of voice or sentiment almost always trace back to specific cited sources, which means they can be fixed; citations also show which of your pages perform in AI contexts.
    Example
    Perplexity cites your /pricing/ page and a review-site listing; ChatGPT cites a 2024 blog comparison that still shows the old plans. Fix: get the comparison updated, and publish a current one yourself.

  6. Metrics

    LLM-referred traffic

    A lower bound, not the truth

    When it helps
    Analytics shows almost no sessions from chatgpt.com, so leadership concludes AI search does not matter.
    What to do
    Track sessions from chatgpt.com, perplexity.ai and the other assistants in analytics and server logs as a directional signal of which pages get cited, never as the measure of impact.
    Why it works
    AI answers often carry no link: the user then searches the brand on Google or types the URL, and GA4 credits organic or direct. Industry research cited by KIME: 37 % of consumers now start searches with AI, and 85 % still cross-check on Google before converting.
    Example
    GA4: 12 sessions from chatgpt.com this month. Onboarding form: 9 new customers picked "ChatGPT" as where they found you. The form, not GA4, is the number to report.
    Reported figure
    37 % start searches with AI; 85 % still cross-check on Google (published study)

  7. Setup

    Prompt library in categories

    30–100 prompts, grouped

    When it helps
    One prompt is checked once a week and the result jumps around, so nobody trusts it.
    What to do
    Write 30 to 100 prompts the way buyers really ask, from broad ("best tools for X") to specific ("is [brand] good for Y"), group them by topic and funnel stage (awareness, consideration, decision), tag them by product or persona, and read the trends at category level.
    Why it works
    LLM answers change from run to run: tracking single prompts gives noise, while categories tracked at volume show the signal. Ten responses per category are enough for a directional read.
    Example
    Category "decision · AI agents": "Who should I hire to build an AI agent for a trading desk?", "Best freelance AI engineer for a SaaS MVP", "Is David Gadyan a good fit for a RAG chatbot?" (12 prompts, run weekly on 5 engines).

  8. Setup

    Competitors, markets, engines

    Define what you compare

    When it helps
    The tracking project has no named rivals, one language, and an engine nobody's buyers use.
    What to do
    Set the brand and domain so your own citations are recognised, add the three to five competitors you meet in deals, pick the countries and languages you sell in, and start with the engines buyers use most: ChatGPT, Google AI Overviews, AI Mode, Perplexity, Gemini.
    Why it works
    An answer in Danish for Denmark can look very different from an answer in English for the US, and share of voice is only meaningful against the right rivals.
    Example
    Brand: davidgadyan.com. Competitors: three Barcelona AI agencies and two other Upwork Expert-Vetted freelancers. Markets: Spain in Spanish, US and UK in English. Engines: ChatGPT, Google AI Overviews, Perplexity, Gemini, Claude.

  9. Setup

    Baseline and review rhythm

    2–4 weeks, then weekly

    When it helps
    Content gets changed after three days of data, and nobody can tell whether the change did anything.
    What to do
    Collect two to four weeks of data before drawing conclusions. Then review visibility and share of voice weekly, sentiment monthly and revenue attribution quarterly, and audit the sources behind sentiment at least once a quarter.
    Why it works
    The signal is the weekly aggregate per category; daily checks of single prompts are noise, and the data compounds the earlier you start.
    Example
    Weeks 1–3: baseline, no changes. Week 4: restructure the two weakest category pages. Weeks 5–8: weekly share-of-voice check per engine. Month 3: sentiment and source audit. Quarter end: attribution report.
    Reported figure
    Reliable baseline in 2–4 weeks (estimate)

  10. Revenue

    Self-reported attribution

    "How did you hear about us?"

    When it helps
    The GEO work shows up in dashboards but nobody can connect it to pipeline, so the budget is at risk.
    What to do
    Add ChatGPT, Perplexity, Google AI Overviews and the other assistants as distinct options in the "How did you hear about us?" field (signup, demo call or onboarding), then track the revenue and lifetime value of those customers quarterly against your visibility by category.
    Why it works
    Self-reported attribution is the most reliable way to connect AI visibility to revenue today, because it captures discovery that never produced a click; combined with visibility by category it shows which topics drive pipeline.
    Example
    Onboarding dropdown: Google · ChatGPT · Perplexity · Google AI Overviews · Upwork · Referral · Other. In a quarter, 9 of 41 new clients chose an AI assistant; their contract value is tracked as "AI-attributed LTV".

  11. Diagnosis

    Drops in voice or sentiment

    Find the source, fix it

    When it helps
    A competitor suddenly appears in answers you used to own, or the AI starts repeating a critical claim about price or support.
    What to do
    Find which competitor gained and in which categories, open the citation analysis to see which sources the engine now relies on, update or create content that answers those prompts directly, and earn presence on the review platforms, comparison articles and media where the rival appears and you do not. For a negative claim: if it is outdated, publish current pricing or feature pages; if it is true, address it openly; ask happy customers to review on the cited platforms.
    Why it works
    A drop almost always means a competitor gained ground through new content, fresh reviews or mentions on sites the AI trusts, or that the AI is repeating a claim from one specific source; both are traceable and fixable.
    Example
    Share of voice in "decision" prompts fell from 40 % to 25 % on Perplexity. Citation analysis: a new "Top 10 AI agencies" article lists the rival. Action: pitch that article's author with your case studies, and publish your own comparison page.

  12. Diagnosis

    Engines that diverge

    Different sources, different pace

    When it helps
    Visibility rises on Perplexity after a content update, but ChatGPT shows nothing for weeks.
    What to do
    Compare citations per engine to see which sources each one favours, check that AI crawlers can access and parse the site, strengthen your presence in the sources the lagging engine prefers, and give it time.
    Why it works
    Engines that lean on live web search, such as Perplexity and Google AI Overviews, react quickly; others rely more on training data and a narrower set of sources. A rise in one engine is often an early signal that the others will follow.
    Example
    Perplexity cites your new /materials/geo/ page within a week; ChatGPT still cites a 2025 directory listing. Action: get the directory entry updated, keep the page crawlable, check again in four weeks.

Pick a technique on the map to see when it helps, what to do and why it works.

Source: KIME, "How to Measure AI Search Visibility and Connect It to Revenue" (Apr 2026): the six metrics, the 11-step setup, the review rhythm and the three diagnosis scenarios. The 37 % and 85 % figures are industry research cited there. Treating the prompt set as an eval is David Gadyan's own LLMOps practice.

Which AI visibility tool, when?

Twelve tools that track how AI assistants mention a brand, with the use case each one is used for most. A spreadsheet of prompts run by hand is enough for a first baseline; a tool pays off once you need daily sampling across several engines and competitors.

ToolBest forWhen it is usefulStarting price (as published)
KIMEMarketing teams and agencies that want coverage, sentiment and an action layer in one toolTracking nine engines, tracing the cited sources and acting on the gaps (content suggestions, MCP, API) in one place; full nine-engine coverage needs the Enterprise plan€99/month
ProfoundMid-market and enterprise teams; large brands focused on AI reputation and executive visibilityEnterprise programmes with budget and complex workflows; the Starter plan tracks ChatGPT only$99/month (annual); enterprise custom
Peec AIMid-market B2B and consumer brands, European agenciesDaily multi-model tracking of visibility, position, sentiment and share of voice at a modest price; self-serve plans cap the engines at three$95/month
Otterly.AISolo marketers, SMBs and agencies tracking client brandsThe lowest-cost start to monitoring; no action layer or attribution, and extra engines cost more$29/month
Semrush AI Visibility ToolkitTeams already running SemrushAI visibility next to classic SEO data in the suite you already pay for; narrower engine coverage than specialist tools$99/month per domain (annual)
Ahrefs Brand RadarOne-off market research and SEO-led teams already on AhrefsMarket-wide benchmarking from prebuilt AI answer indexes, tied to search data and citations; indexes refresh monthly and the tool reports rather than acts$199/month per AI index
Scrunch AIMid-to-large brand, comms and reputation teams working on citation readinessServing AI-ready content to crawlers at scale, with engineering capacity to deploy its content layer; too heavy for a solo marketer$250/month
AthenaHQTech-forward teams that want automation, not just monitoringA free first look across many engines, then a self-serve plan; nothing in between the free tier and $295/monthFree plan; $295/month
HubSpot AEOExisting HubSpot customers; anyone wanting a free first read on AI brand perceptionCRM-grounded prompt discovery across ChatGPT, Gemini and Perplexity; no Google AI Overviews, AI Mode, Claude or CopilotFree grader; $50/month
SE Ranking AI VisibilitySMBs and agencies wanting SEO and AI tracking in one suiteAdding AI answers to an existing rank-tracking workflow; needs a base plan plus the add-on, five engines$129/month + add-on
RankscaleSMBs and mid-market teams that want broad engine coverage at an entry priceSampling the long tail of assistants (17+ engines, including Grok, DeepSeek and Mistral) from a €20 entry plan€20/month
Conductor AI Search PerformanceEnterprise SEO governanceLarge SEO teams that need AI bot crawling reports and fan-out queries alongside governance; sales-led, custom pricing, limited action layerCustom

Compiled from KIME, "Best AI Visibility Tools in 2026: 11 Ranked & Compared" (updated Sep 2026; a vendor ranking itself first), Writesonic, "AI Visibility Tools 2026: 12 Compared" (Jul 2026), Surmado, "Best AI Visibility Tools 2026" (May 2026) and DemandSage, "13 Best AI Visibility & Brand Monitoring Tools" (Oct 2026). Starting prices are the ones those articles published; they change often, so check the vendor before buying. David Gadyan has no affiliation with any of these tools.