# AI Visibility Metrics and KPIs: What to Track and Report The AI visibility metrics and GEO KPIs worth tracking, how to calculate each without inflating it, and which to report to leadership. **Published:** November 5, 2026 **Author:** Zain Zia & Connor Lahey --- If you’ve spent years reporting rankings and organic traffic, AI visibility metrics can feel like learning the job again. There’s no fixed position one, and the same prompt can change answers an hour later. And, worst of all, some AI visibility tools show every number they track but don’t tell you which ones matter or what to do when one drops. In this guide, we cover the nine AI visibility metrics worth tracking, how to calculate each one without inflating it, and how to distinguish a real performance drop from normal run-to-run variation. We also show which KPIs to report to leadership, which to work on weekly, and how to set targets without a made-up benchmark. **Key takeaways** - AI visibility metrics show whether buyers asking about your category on ChatGPT, Perplexity, or Google’s AI Overviews ever see your brand, get linked to your site, or read accurate answers about you. - The four metrics worth setting targets against are presence rate, share of voice, citation rate, and AI referral traffic. - Every AI visibility percentage needs a stated denominator, and prompts that name your brand should be reported separately, because both can make your headline number look far better than your real visibility. - A rise or drop is only worth acting on once it holds across several runs and several weeks, and checking whether one engine or all of them moved shows where the cause lies. - Leadership needs a few KPIs once a month, with the reason behind each move; the team needs prompt-level numbers every week; and targets should come from your own baseline and the competitor ranked just above you. - Searchable tracks these metrics across the major AI engines, labels the prompts that name your brand so you can filter them out, ranks your share of voice against competitors, brings in AI referral and crawler data through LLM Analytics, and turns it all into shareable reports. ## What are AI visibility metrics and KPIs? **AI visibility metrics** measure how often, how prominently, and how accurately AI answer engines such as ChatGPT, Perplexity, and Google’s AI Overviews mention or cite a brand across a fixed set of prompts. **AI visibility KPIs** are a subset of those metrics a team sets targets against, such as presence rate, share of voice, and citation rate, plus a downstream business outcome like AI referral traffic. The two terms aren’t interchangeable, though. A team might track dozens of metrics but agree to be judged on only a few of them. However, the other metrics still matter because they help explain why one of those KPIs moved. Every AI visibility metric is calculated from a prompt set, which is the list of questions a team chooses to track. Take a project management tool, for example, where buyers can ask AI engines hundreds of questions about the category but only some of them matter to the business, such as “best project management software for a small agency” or “how do I manage client projects without email.” Your presence rate or share of voice only describes your visibility on the questions in that set. So if the set leans on definitional prompts like “what is a Gantt chart” and leaves out comparison prompts like the agency one, a strong presence rate can hide the fact that buyers comparing vendors aren’t seeing you. Together, these metrics show the results of your [answer engine optimization](https://www.searchable.com/blog/what-is-aeo) (AEO) work. You’ll also hear the term generative engine optimization (GEO), and teams that use it call these numbers GEO KPIs. Others describe them as AEO metrics or LLM visibility metrics. ## How do you calculate AI visibility metrics? AI visibility metrics are calculated from a set of tracked prompts, each run several times on each AI engine, so every AI visibility percentage needs a stated denominator, usually prompts or responses. The same data can produce very different numbers depending on which one you divide by. In short, what you divide by matters as much as what you count. If you’re not sure where your tool’s responses come from in the first place, check out our guide on [AI visibility tracking](https://www.searchable.com/blog/ai-visibility-tracking). ### Per prompt or per response: which denominator should you use? Take, for example, a brand that tracks 50 prompts, with each one running four times on three engines every week. That’s 600 responses a week, and say the brand shows up in 150 of them. Count per response, and the brand’s visibility is 150 out of 600, or 25%. In other words, on any single run of any single prompt, the brand turns up about a quarter of the time. Count per prompt, and the number looks very different. Each prompt gets 12 responses a week, and the brand’s 150 mentions are spread across 32 of the 50 prompts. Because the brand appears at least once for 32 prompts, its per-prompt figure is 64%. The per-prompt figure will never be lower than the per-response one, because one mention is enough for a prompt to count, while every response counts on its own. ![Diagram of 50 prompts run 4 times on 3 engines, giving 600 responses, showing the same data as 25% per response and 64% per prompt](https://www.searchable.com/blog/ai-visibility-metrics-denominator-diagram.webp) So pick one denominator, label it in every report, and keep it the same from one period to the next. If you switch from per response to per prompt halfway through a quarter, visibility can jump from 25% to 64% when the only change is the math behind it. ### Why prompts that name your brand inflate AI visibility metrics Once you’ve picked a denominator, there’s a second thing to check: which prompts you’re counting in the first place. Some prompts in your set will probably name your brand outright, like “Is [Brand] good for remote teams?” or “[Brand] vs [Competitor].” Your brand appears in nearly every answer to those, since the question already positions it there. The trouble starts when you mix these branded prompts in with the rest of your prompts, because they push the headline number up without telling you whether new buyers are finding you. Let’s go back to the brand from the earlier example and say 10 of its 50 prompts name it explicitly, and it appears for all 10. Those 10 prompts make up nearly a third of the 32 where the brand shows up. If you take them out, the brand appears for 22 of the remaining 40 prompts, so its per-prompt figure drops from 64% to 55%. ![Grid of 50 prompts showing 10 that name the brand, 22 that do not but still return it, and 18 where it never appears, with presence falling from 64% to 55% once the 10 are removed](https://www.searchable.com/blog/ai-visibility-metrics-branded-prompts-diagram.webp) Sure, nine points doesn’t sound like much, but real tracking data shows a far bigger gap. Across brands tracked in Searchable in September 2026, the median brand was named in 93% of answers to prompts that named it and in only 7% of answers to prompts that didn’t. This is why it’s worth separately reporting on prompts that don’t name your brand and treating that number as your real visibility figure. Keep tracking the brand-name prompts too, though, because they show how AI engines answer people who are already considering you. Here’s how that split looks in practice, with one brand’s visibility in Searchable filtered by branded and unbranded prompts over the same week. ![Two Searchable Visibility charts for the same week, with Branded only at a 95.9% visibility score and Unbranded only at 21.1%](https://www.searchable.com/blog/ai-visibility-metrics-branded-vs-unbranded.webp) ## The AI visibility metrics worth tracking Nine AI visibility metrics cover what most teams need, in four groups: - **Presence:** Presence rate, share of voice, average position - **Sources:** Citation rate, mentions without a citation - **Answer quality:** Sentiment, accuracy - **Business outcomes:** AI referral traffic, branded search demand Below, we’ll look at the formula behind each metric and the decision it drives. ## Presence The first three metrics show whether AI engines mention your brand at all and how you compare with competitors when they do. ### Presence rate Presence rate is the share of tracked prompts where your brand appears at least once. Here’s how to calculate it: **Presence rate = prompts where the brand appears at least once ÷ prompts tracked** So if you track 50 prompts and your brand shows up in answers to 15 of them, your presence rate is 30%. Each prompt counts once, whether your brand appeared on a single run or on every run for that prompt. Two things tend to increase presence rate: publishing content on topics you hadn’t covered before and getting your brand mentioned on other sites. If it falls, you can reopen the prompts you dropped out of and see what the engines say now. Those answers show which competitors and sources took your place, so you know where to focus first. ### Share of voice in AI answers Share of voice in AI answers compares how often answers name your brand with how often they name the competitors you track. To work it out, divide your mentions by everyone’s mentions: **Share of voice = mentions of your brand ÷ mentions of all tracked brands, including yours** If AI answers mention your brand 120 times in a week and your tracked competitors 360 times between them, your share of voice is 25% (120 out of 480). One thing to watch out for here is that share of voice is only reliable when your prompt set and competitor list stay the same from one reporting period to the next. Add a competitor or swap out a few prompts, and the number will move for reasons unrelated to your performance, so lock both before you compare periods. And since it’s a ratio, share of voice also moves when competitors gain or lose mentions, even if your own count doesn’t change. If it falls, you can check which side of the ratio changed. Fewer mentions of your brand point back to your own content, while more mentions of a rival point to wherever that rival is getting cited. In Searchable, for example, the share of voice view ranks your brand against its top competitors for each day, so you can see who gained ground when your share moves. ![Searchable Share of Voice chart ranking a brand against its top 5 competitors for each day, with competitor names blurred](https://www.searchable.com/blog/ai-visibility-metrics-share-of-voice.webp) ### Average position in AI answers Average position shows where your brand tends to appear within an answer, whether it’s the first brand named or further down the list. If engines usually mention you first, your average position sits close to one. Here’s how to work it out: **Average position = sum of the brand’s positions in answers that name it ÷ answers that name it** Say your brand shows up in four answers this week. It’s named first in two of them, then third in one and fourth in the other, so its average position is 2.25, or (1 + 1 + 3 + 4) ÷ 4. Lower is better here, so if that number rises to 2.5, you’ve slipped down the list. Averaged across many runs, the number shows whether engines tend to lead with you or mention you late. Always read it alongside presence rate, because answers that don’t name you aren’t counted at all. A brand that appears for only 2 of its 50 tracked prompts, and is named first every time it does, averages position 1. Its average position looks perfect, but its presence rate is just 4%. Average position usually shifts when engines start listing other brands ahead of yours, or stop doing so. If your position falls while presence rate holds, you’re still in the answers, just further down the list. So look at how engines describe the brands that now come before you, and note what they say about those brands that they don’t say about you. ## Sources The next two metrics look at the links in AI answers and whether they point to your own site. ### Citation rate Citation rate is the share of AI answers that link to a page on your own domain as a source. It’s easy to confuse with a mention, but the two aren’t the same. A mention means the answer names your brand, while a citation means it links to one of your pages as a source, and an answer can do one without the other. Here’s how to calculate it: **Citation rate = responses that link to the brand’s own domain ÷ responses** Say your prompts produce 400 responses this week and 60 of them link to a page on your site, so your citation rate is 15%. Seeing [mentions and citations](https://www.searchable.com/features/aeo-insights/mentions-citations) side by side makes the gap easier to spot, since a brand can be named often and still linked rarely. Citation rate tends to move when you publish or update pages on the topics your prompts cover, or when engines start citing different sites for those topics. If it falls, you can check which of your pages stopped being cited. A page that was linked last month and isn’t now is usually the best place to start, because you already know it can earn a citation. ### Mentions without a citation This metric shows how often AI answers talk about your brand without linking to your site to back it up. It isn’t simply the opposite of citation rate, because it only counts answers that name you (not whether they cite you when doing so). Here’s the formula: **Mentions without a citation = responses that name the brand but link to no page on its domain ÷ responses that name the brand** Say 80 answers name your brand this week, and 52 of them don’t link to any page on your site. That’s a rate of 65%. A high rate means engines know who you are but rely on other sites, such as review platforms, comparison articles, or forums, to describe you. In other words, what buyers read about you comes from someone else’s page. The rate tends to drop when your own pages start getting cited on prompts where you’re already named, and it can climb when coverage of your brand on other sites grows faster than your own. If it climbs, look at which [citation sources](https://www.searchable.com/features/aeo-insights/sources) those answers link to instead, then work on getting your brand featured on those pages. That could mean a listing on a review site or a spot in a comparison article. ## Answer quality Being named is only part of the picture, so the next two metrics check what AI answers say about you once you’ve earned mentions. ### Sentiment in AI answers Sentiment tracks whether AI answers describe your brand positively, neutrally, or negatively when they mention you. Each tone gets its own share: **Sentiment share = mentions with a given tone (positive, neutral, or negative) ÷ all mentions of the brand** If engines mention you 200 times in a month and 30 are negative, 15% of your mentions are negative. Tools usually handle [sentiment tracking](https://www.searchable.com/features/aeo-insights/sentiment) automatically, so sarcasm, side-by-side comparisons, and mixed answers can be misread. That’s why it’s worth inspecting a handful of scored mentions manually each month, starting with the negative ones, before you report the number anywhere. Sentiment tends to shift alongside new reviews, news coverage, or forum discussions about your brand. If your positive share falls, read the negative mentions to see sticking points customers keep coming back to, such as pricing, support, or a missing feature. A recurring theme tells you whether the fix is a product change, a support change, or a content change. ### Answer accuracy Answer accuracy measures how often AI answers get a verifiable fact about your brand wrong, such as a price, a feature, or whether a product is still available. You calculate it manually from a sample of answers: **Answer error rate = sampled answers with at least one wrong fact ÷ sampled answers** Let’s say 40 answers mention your brand; you check each one against your current pricing and product pages, and find four with at least one mistake. Your error rate is 10%. Answer accuracy is usually checked by hand via a monthly sample, since a tool can only flag a wrong price if it already knows the right one. Keeping a short list of frequently changing facts, like prices, plan names, and integrations, makes each check quicker. Errors tend to show up after you change pricing, rename a plan, or retire a product. If your answer error rate climbs, note which facts are wrong and which pages the answers cite. Update your own pages first, then ask the owners of outdated pages on other sites to correct them. ## Business outcomes The last two metrics show whether AI visibility is translating into real business impact, both on your site and in search. ### AI referral traffic AI referral traffic counts visits to your site from AI assistants, such as ChatGPT, Perplexity, Gemini, and Claude, as recorded in Google Analytics 4. Track it as a share of your total traffic: **AI referral share = sessions from AI assistants ÷ all sessions** If your site gets 50,000 sessions in a month and 1,000 of them come from AI assistants, AI referrals make up 2% of your traffic. The number undercounts, though, because many people read an AI answer and never click through to the sources. AI-referred visitors also tend to arrive further along in their research, so it’s fairer to judge the channel by what each visit is worth, such as sign-ups or demo requests, than by its visit count. AI referral traffic tends to follow citation rate, since a visitor can only click through on an answer that links to your site. If it falls, check your citation rate for the same period. A drop in both points to lost citations, while a traffic drop on its own is worth checking against your analytics setup before you act on it. ### Branded search demand Branded search demand tracks how many people search for your brand by name in Google, using impressions from Search Console. Since the overall trend is what’s most telling, measure it against the prior period: **Branded search growth = (branded impressions this period − branded impressions last period) ÷ branded impressions last period** Filter Search Console’s Performance report to queries that contain your brand name. If those impressions go from 8,000 one month to 9,000 the next, branded search grew 12.5%. A branded search demand increase that follows a rise in AI visibility supports the case that AI answers are sending people to look you up, but it doesn’t prove it. Ads, press coverage, a product launch, or seasonal demand can lift branded searches just as easily. If it falls while presence rate and share of voice hold steady, the cause probably sits outside AI search, so look at your other channels first. If branded search demand falls alongside them, the drop in AI visibility is the more likely place to start. ## Leading and lagging AI visibility KPIs AI visibility KPIs fall into three layers that move at different speeds: inputs such as whether AI crawlers can reach your pages, visibility metrics such as presence rate and share of voice, and business outcomes such as AI referral traffic and pipeline. Inputs and visibility metrics usually respond before business outcomes do, so judging a program on outcomes alone can end good work before it has had time to show up. Here’s how the three layers line up: | Layer | Metrics in it | How quickly it typically responds | | :---------------- | :-------------------------------------------------------------------------------------------------------------------------------------- | :---------------------------------- | | Inputs | AI crawler visits (tracked with [LLM Analytics](https://www.searchable.com/features/llm-analytics)), whether your key pages can be crawled | Fastest, often within days of a fix | | Visibility | Presence rate, share of voice, average position, citation rate, mentions without a citation, sentiment, answer accuracy | Slower, shifting over several weeks | | Business outcomes | AI referral traffic, branded search demand, pipeline (see how they connect across the [AI marketing funnel](https://www.searchable.com/blog/ai-marketing-funnel)) | Slowest, often months | Reading the layers together also shows you where a program is stuck. For example, if crawler visits climb but presence rate hasn’t moved after several weeks, engines can reach your pages, so the next thing to look at is what those pages say. And if visibility is rising while referral traffic hasn’t yet, you’re most likely just early, since the business outcomes layer is the last to respond. ## How to tell whether a change in an AI visibility metric is real A change in an AI visibility metric is only worth acting on when it holds across several runs and several weeks, because AI answers vary from one run to the next even when nothing has changed. [A 2026 State of AI Search report](https://www.airops.com/report/the-2026-state-of-ai-search) found that “only 30% of brands stay visible from one answer to the next, and just 20% remain present across five consecutive runs.” So before you act on a rise or a drop, run it through three rules: 1. **Keep the prompt set fixed.** Adding or removing prompts changes every number in your report at once, so treat any change to the set as the start of a fresh baseline. 2. **Compare four-week trends with earlier four-week trends.** A single week can swing on run-to-run variation alone, so set the last four weeks against the four before them before you call anything a trend. 3. **Check whether one engine or all of them moved.** A drop on a single engine and a drop across every engine point to different causes, which the table below helps you tell apart. Once a change passes all three, here’s how to read what it usually means: | What you see | What it usually means | | :-------------------------------------- | :-------------------------------------------------------------------- | | One engine fell, the others held | That engine changed something, and your site probably isn’t the cause | | Presence rate flat, share of voice down | A competitor gained ground, while your own visibility held | | Presence rate down, share of voice flat | The whole category got quieter, often for seasonal reasons | | Both down, concentrated in one topic | Your content in that topic has a problem | For a full walkthrough, check out Searchable’s guide to [diagnosing a visibility drop](https://docs.searchable.com/recipes/visibility-drop). ## Which AI visibility KPIs to report to leadership, and which to work weekly Leadership needs three or four AI visibility KPIs once a month, with the trend and one sentence on what caused it. The team doing the work needs prompt-level and topic-level numbers every week because they show which content to change. Here’s how that split looks in practice: | Audience | KPIs | Cadence | The question it answers | | :--------- | :--------------------------------------------------------------------------------------------------------------------- | :------ | :------------------------------------------------------------------------------- | | Leadership | Presence rate on prompts that don’t name the brand, share of voice rank against named competitors, AI referral traffic | Monthly | Are more new buyers finding us in AI answers, and are we gaining on competitors? | | Team | Presence by topic and prompt, citation rate, mentions without a citation, sentiment | Weekly | Which content should we change next? | | Team | Answer accuracy | Monthly | Are AI answers getting our facts right? | Leadership cares more about why a number changed than the number itself, so include the answer in the same update as the KPIs. You could write something like: “Presence on prompts that don’t name us rose from 18% to 22% this month, mostly in the integrations topic, after we published three comparison pages.” Keep a running log of every page you publish or update throughout the month, plus any competitor launches, pricing changes, or campaigns you track. When a number shifts, check the log first, because the likely cause is often already in it. This will help you provide your leadership with accurate insights. And since the report follows the same format every month, you can also [automate reporting](https://www.searchable.com/workflows/automate-reporting) and spend your time deciding what to do next rather than pulling the numbers together. ## How to set targets for AI visibility KPIs Set AI visibility targets from your own baseline and your competitors’ share of voice, because no universal benchmark exists. The number depends on which prompts you track, which competitors you compare against, and which engines you include. So instead of borrowing a target from someone else’s dashboard, build one from your own numbers in three steps: 1. **Record four weeks of baseline.** Track your KPIs for four weeks before you start making changes, so your starting point reflects a typical month rather than one unusually good or bad week. 2. **Find the share of voice gap to the competitor directly above you.** That brand is the most realistic one to catch, and the gap between your share of voice and theirs is the ground you need to make up. 3. **Set a target to close part of that gap in a quarter.** Aiming to close all of it in three months usually sets the team up to miss, so pick a share of the gap you can defend. Say your baseline share of voice is 12% and the competitor ranked just above you sits at 18%. A target of 15% by the end of the quarter closes half the gap, and it’s a number you can explain to anyone who asks where it came from. ## How Searchable tracks AI visibility metrics Searchable calculates most of the metrics in this article from a fixed set of tracked prompts run across up to nine AI engines, and can split every metric by engine, topic, and prompt. Prompts that name your brand carry a “Branded” label, so you can report them separately without inflating presence rate and citation rate. So here’s roughly how your first month would go: 1. **Set up your prompts.** Head to [**“Prompt Research”**](https://www.searchable.com/features/prompts) and add the prompts your buyers would ask, either one at a time with **“Add Prompts,”** in bulk with **“Bulk Upload,”** or by letting Searchable suggest them with **“Generate Prompts & Topics.”** Your prompts then run daily, and you can see each one’s results in **“Your Prompts,”** where prompts that name your brand show up with a “Branded” label. ![Searchable Prompt Research page with the Quick Prompt Add dialog open, showing the Add Prompt and Bulk Upload tabs](https://www.searchable.com/blog/ai-visibility-metrics-add-prompts.webp) 2. **Split branded prompts.** Use the **“Branded & Unbranded”** filter at the top of AEO Insights to switch to **“Unbranded only”** or **“Branded only,”** so you see every number both ways. 3. **Check where you stand.** [AI visibility analytics](https://www.searchable.com/features/aeo-insights/visibility) in AEO Insights then give you: - Your visibility score, average position, and sentiment - Share of voice, ranked against the competitors you track - Separate tabs for mentions and citations, and for the sources AI answers cite - Filters for model, topic, and location on every view ![Searchable Mentions and Citations chart showing daily mentions far above daily citations over 30 days](https://www.searchable.com/blog/ai-visibility-metrics-mentions-citations.webp) 4. **Connect your traffic data.** LLM Analytics pulls AI referral visitors for each assistant from GA4. Once you connect your hosting platform logs, it also shows AI crawler visits. 5. **Share a report.** Reports turn the month’s numbers into a summary you can send to leadership or a client as a published link, by email, or as a PDF. If you’re an agency, you can also white-label it. ![Searchable report editor with the Share report panel open, showing Publish, White-label, Share via email, and Download PDF options](https://www.searchable.com/blog/ai-visibility-metrics-share-report.webp) By the end of the month, you have everything you need for a baseline. Note your visibility score with the filter on **“Branded & Unbranded,”** then switch to **“Unbranded only”** and check how far it drops. Then use the share of voice rankings to set your first target, following the steps above. ## Frequently asked questions These answers cover what comes up once the core metrics are in place: how metrics differ between engines, whether to track each market separately, how AI share of voice compares with SEO share of voice, and whether any of this works without a tool. ### Do AI visibility metrics differ between ChatGPT and Google AI Overviews? Yes, each engine cites sources differently, so the same brand can look strong in one and barely show up in another. That’s why it’s worth reporting each engine separately first, then combining them. Otherwise, a gain on one engine can hide a drop on another, and your blended number looks steady on the surface while one engine is slipping underneath. ### Should you track AI visibility metrics separately for each country? Yes, if you sell in more than one market. Answers to the same prompt can differ by location, so a brand that leads in the US might barely appear in the UK. The safer approach is to fix the location for each prompt set, then [report each market separately](https://www.searchable.com/features/aeo-insights/location) rather than averaging them. ### Is AI share of voice the same as SEO share of voice? No, they measure different things. SEO share of voice is usually estimated from your rankings and the search volume behind each keyword. AI share of voice, on the other hand, counts how often answers to a fixed set of prompts mention your brand, compared with your competitors. This means a brand can rank well in Google and still be left out of AI answers, or vice versa. ### Can you track AI visibility metrics without a tool? For a handful of prompts, yes, by hand. But the work adds up quickly. For example, tracking just 20 prompts, run 4 times on 3 engines each, gives you 240 answers to read and log every week, and each one needs checking for mentions, position, citations, and tone. Past a few dozen prompts, it gets hard to keep up by hand. ## Set your AI visibility baseline this week with Searchable If you do one thing this week, make it this: choose the prompt set you’ll track, split out the prompts that name your brand, and record a baseline before you change any content. Every target and leadership report you build later depends on that starting point. And if you’d like a head start, Searchable’s free visibility report gives you a first read on where your brand shows up in AI answers today, so your baseline starts from real numbers. [**Get your free visibility report →**](https://www.searchable.com/free-visibility-report) --- [Back to Blog](https://www.searchable.com/blog) | [Searchable Homepage](https://www.searchable.com)