Someone on your leadership team asked what your AI search strategy is. You opened ChatGPT, typed your core use case, and watched it name a competitor instead of you.
That stings a little. It also means you're paying attention at exactly the right time.
Here's the good news. AI visibility measurement is learnable, and you can start this week with tools you already have. This guide gives you the full mental model: what to measure, how to run your first audit, how to benchmark against competitors, how to set up ongoing tracking, and how to prove the work is moving the numbers. If you've been asking how to track brand in AI answers without a data team or a big budget, this is the map.
Think of it as your AEO measurement guide, one hub that frames the whole program and points you to the deeper pieces when you need them. You don't need to measure AI search visibility perfectly on day one. You need a baseline and a next step. Let's build both.
Why AI visibility measurement is its own discipline now
Your buyers are getting answers inside AI, not just on a results page.
ChatGPT reached around a billion monthly users by mid-2026 and now handles close to 18% of all digital queries. Google's global search share slipped to its lowest point in a decade. AI Overviews show up on roughly one in six searches, and in some verticals closer to one in four. When an AI Overview appears, far fewer people click a traditional result at all.
Your old SEO dashboards can't see any of this. They count rankings and clicks on a page. They cannot tell you whether an AI answer named your brand, linked your page, or handed the moment to someone else.
That gap is why AI visibility measurement exists as its own practice, a close relative of answer engine optimization. It measures presence inside the answer, not position on the page. Different signals, different cadence, different playbook. To measure AI search visibility, you watch what the model says about you, not where a blue link lands.
Your first action here is simple. Open ChatGPT, Perplexity, and Google's AI Mode, and ask each one three questions a buyer would actually ask about your category. Note whether you appear, whether you get linked, and how the answer describes you. That rough snapshot is your starting line, and it's the seed of everything that follows.
What "AI visibility" actually means
AI visibility is your measurable presence inside answers generated by AI search and assistants. It breaks into four signals, and keeping them separate keeps you honest.
- Mention: your brand name appears in the answer, with or without a link.
- Citation: the AI lists a URL you own as a source.
- Position: where you land when several brands get listed in one answer.
- Sentiment: the tone the answer uses around your name.
These four are not the same thing, and confusing them is the most common early mistake. A model can mention you warmly without linking you. It can cite your page while listing you third. Understanding how citations and brand mentions differ is worth the ten minutes it takes, because you optimize each one differently.
Both matter. A citation earns you a click. An unlinked mention still shapes what the buyer believes before they ever reach your site. Learning to read linked and unlinked mentions together gives you the full picture.
Your action here: pick one buyer question and, for each answer you get, label all four signals by hand. Do it five times and the definitions stop being abstract.
The metrics that matter, and how to read them
You don't need every metric. You need a small set you can track consistently. Here are the ones that carry weight.
- Mention Rate: the share of your tracked prompts where the AI names your brand.
- Citation Rate: the share where the AI links to a page you own.
- Share of Voice: your mentions divided by all brand mentions across your tracked competitor set.
- Average Position: where you typically land when you are listed among others.
- Net Sentiment Score: positive minus negative mentions over total, on a scale from minus 100 to plus 100.
- Prompt Coverage: how many of your tracked prompts show you appearing at all.
- AI Referral Sessions and Conversions: visits and conversions from AI referrers, pulled from your analytics.
What counts as good? There's no published industry standard yet, so treat these as working rules of thumb, not laws. Practitioners tend to read a mention rate above 50% as dominant, 20 to 50% as visible but not the default answer, and under 20% as under-represented. Your own trend matters more than any absolute number.
Two metrics deserve special care. Share of voice tells you how you stack up against rivals on the same questions, so it's your competitive scoreboard. Sentiment is the one teams skip and regret, because a high mention rate wrapped in negative framing is worse than a smaller one wrapped in trust. If you only add one metric beyond mentions this month, make it sentiment. Start with the KPIs that matter and ignore the rest until these feel routine.
One more worth watching over time is source diversity, the count of distinct third-party domains that mention or cite you. Visibility built on a single Reddit thread or one trade publication is fragile, because it collapses the moment that source shifts. A healthy spread of sources is a sign your presence is durable, not lucky. You don't need to chase it on day one, but keep an eye on it as your program matures.
Which AI engines to track first
Start with the engines your buyers actually use, then widen.
For most B2B and SaaS teams, the minimum viable set is ChatGPT, Perplexity, and Google's AI Mode or AI Overviews. Add Gemini and Claude once you have a baseline. Beyond that, Copilot, Grok, and Meta AI are worth watching for reach, not obsessing over on day one.
Why not track everything at once? Because each engine cites differently, and spreading yourself thin across ten of them buries the signal. ChatGPT often answers from body text without listing sources, so on that engine you're mostly measuring mentions, not links. Perplexity visits many pages per query and cites a handful, so citations there are gold. Google's AI Overviews link a source in the large majority of responses, which makes them a citation-heavy surface worth watching closely. Gemini and Claude each have their own quirks that only show up once you're tracking them. A win on one engine can be near-invisibility on another, so you want depth on a few before breadth across many.
The tiering many teams follow maps neatly to budget: ChatGPT alone to start, then Perplexity, then Gemini, then the full set once the program earns it. That's the same ladder most trackers price against, so your engine coverage and your spend tend to grow together. Your action: commit to two engines this week. Two tracked well beats five tracked never.
How to run your first AI visibility audit
An audit is just a structured look at where you stand today. You can do a useful one in about 30 minutes.
Here's the shape. Spend five minutes on scope: pick your two or three engines and the handful of topics that matter most. Spend ten building a small prompt library, the real questions buyers ask. Spend ten running each prompt on each engine and capturing what comes back. Spend the last five scoring: did you get mentioned, cited, and in what tone?
That's the whole thing. It's not elegant and it doesn't need to be. The point is a baseline you can repeat. When you're ready to go deeper, a full step-by-step audit of your brand's presence in AI answers walks through capture and scoring in detail. From there, a longer visibility audit plan turns the snapshot into a repeatable cycle. That recurring approach is exactly what established AEO guides recommend, so you're not inventing a method from scratch.
One reassurance before you start. Your first audit will feel messy and incomplete. That's normal. A rough baseline you actually finish beats a perfect one you never start. And once you've done it once, you know how to track brand in AI answers well enough to repeat it every week without thinking twice.
Build a prompt library that reflects real buyers
Every metric you track is only as good as the prompts behind it. This is the step teams rush, and it's the one that decides everything downstream.
Mine real buyer language, don't invent it. Pull questions from sales call notes, support tickets, your Search Console queries, Reddit threads, and competitor reviews. The words buyers actually type are rarely the words you'd write in a brief.
Then cover the whole journey:
- Awareness: "what is [category]"
- Consideration: "best [category] for [use case]"
- Comparison: "[you] vs [competitor]"
- Decision: "is [product] worth it," "[product] pricing," "[product] reviews"
- Implementation: "how to use [product] for [job]"
Include competitors by name, because "X vs Y" is where a lot of decisions actually get made. Write two or three phrasings of each prompt, since small wording changes expose how differently engines respond. Aim for at least 50 prompts before you trust a weekly number. Fewer than that and the noise drowns the signal.
Your action: write ten real buyer questions today, in the buyer's words, not yours. You'll expand later. Ten honest prompts beat 50 invented ones.
Benchmark: what "good" actually looks like
Here's a freeing truth. There is no public benchmark for AI visibility, so you can stop hunting for the magic number. You benchmark against three things instead.
First, your own history. Last week versus this week is the trend that tells you whether the work is landing.
Second, named competitors on the exact same prompt set over the exact same period. This is the fair fight, and it's where share of voice earns its keep.
Third, the category floor: roughly what a generic, unoptimized brand gets for free. Anything above that floor is a real result.
Remember that share of voice scales with the size of your competitive set. Thirty percent share in a five-brand race makes you the leader. The same 30% in a twenty-brand field puts you mid-pack. Always read the number against the crowd it came from. If you want the full method, benchmarking your AI visibility against competitors breaks it down step by step.
Your action: pick your three closest competitors and run your prompt set against them once. That single comparison reframes everything.
What tools you actually need to track AI citations
You can start by hand, and for a small program you should. But at some point you'll want tools that track AI citations for you on a schedule, so you're reviewing numbers instead of copying answers into a spreadsheet.
Three layers make a complete stack. First, an AI visibility tracker that runs your prompts across engines and reports mentions, citations, and sentiment. Second, an SEO platform for the keyword and backlink context underneath, since AI engines cite the open web. Third, web analytics to catch referral traffic. Most teams already own the second and third. The tracker is the new purchase.
The tracker market is young and moving fast, so prices and features shift every quarter. At the affordable end, monitors built for solo marketers and small teams start around $30 per month for a handful of prompts. Mid-market platforms land in the $100 to $500 range and add competitor benchmarking, sentiment, and daily refresh. Enterprise platforms run higher and quote-based, adding crawler-level analytics and agency workspaces. Established SEO suites now bundle AI visibility too. Ahrefs Brand Radar tracks share of voice across major models using a large index of AI prompts. The Semrush AI Visibility Toolkit adds prompt tracking and sentiment inside a dashboard you may already use.
Which do you need? If you're just proving the case to leadership, a cheap monitor or a free grader is plenty. If AI search is becoming a real channel, you want something that tracks AI citations across several engines and benchmarks you against competitors on the same prompts. A closer look at the AEO tools built to track AI citations compares the options in depth, so you can match spend to stage. Most vendors are transparent here too, and Semrush lists its AI-search plan pricing openly if you want to size the cost first.
One caution before you buy. Definitions differ between tools. Some count a mention only in the first 200 characters of an answer, others count any appearance anywhere. Don't compare absolute numbers across two tools without checking how each one defines a mention. Read the trend inside one tool, not the gap between two.
Set up ongoing tracking and keep the data clean
A one-time audit tells you where you are. Ongoing tracking tells you whether you're winning. The difference is a few habits.
Pick a cadence that fits your category. Weekly for fast movers like SaaS and agencies, where competitors ship constantly. Bi-weekly or monthly for slower fields. And check the engines themselves each quarter, because a model update can shift your numbers without you changing a thing.
Then protect your trend line with a little discipline:
- Lock your prompt library once it stabilizes. Adding or dropping prompts resets your trend. If you must grow it, version it clearly as a v2 set rather than editing the original.
- Run each prompt several times per cycle. Model output varies run to run. Research on repeated prompts found almost no chance that two runs return the same brands in the same order, and only around 60% overlap between runs. So report every metric as an average, never a single reading.
- Keep your sample honest. Fifty prompts is a floor for stable weekly numbers, 100 is comfortable, 200 is enterprise-grade.
One more thing worth knowing. Most tools query the model APIs rather than the chat window, so their results differ slightly from what a logged-in user sees, because APIs skip personalization. This is also why the same brand can look different across two trackers, and why AI citations keep changing even when your content hasn't. It's not a bug in your program. It's the nature of the medium.
Your action: put a 30-minute tracking review on your calendar, same day each week. Consistency is the whole game.
Prove your AEO work moved the numbers
This is the hardest part of AI visibility measurement, and the part your leadership will ask about first. Be rigorous and be honest, in that order.
Work up three levels of proof:
- Correlation reporting. Track mention rate, citation rate, and share of voice over time next to AI referral sessions and AI-assisted conversions from your analytics. Show the lines moving together. Report correlation, not causation. This is the minimum bar every program should clear.
- Branded-search lift as a proxy. When AI mentions rise, branded searches tend to follow within a few weeks. Watch your branded impressions and clicks in Search Console as a lagging sign that mentions are landing.
- Holdout testing. The cleanest proof. Leave one matched set of prompts alone as a control, intervene on another, and measure the difference. It's rarely done well, but it's the gold standard when the stakes justify it.
Now the honesty part. The direct line from "saw your brand in ChatGPT" to "signed the contract" is still mostly dark. It's multi-touch and rarely self-reported, and no tool can yet draw it cleanly. Say so plainly. Over-promising on attribution is how AEO programs lose credibility with the exact leaders they're trying to win over.
The simplest report that works: three numbers, monthly. AI referral sessions, branded-search lift, and mention or citation rate. When all three move in the same direction, your program is working. When they diverge, that's a signal to dig, not a reason to panic. Maybe a model update shuffled citations, or a competitor shipped a page that's now winning your prompts. The trend tells you where to look next, which is exactly what a measurement program is for.
The measurement mistakes to avoid
Most teams trip on the same few things. Knowing them in advance saves you months.
- Trusting a single run. Variance is real. Always average several runs.
- Optimizing one engine. A winner on ChatGPT can be invisible on Perplexity. Read them together.
- Ignoring sentiment. Being mentioned a lot with negative framing is a problem, not a win.
- Changing prompts mid-program. It quietly invalidates your trend. Version instead.
- Confusing rank with mention. Being listed first in one answer is not the same as appearing across many. Track both.
- Measuring for the dashboard. These numbers exist to guide better content, not to score a game.
None of these are hard to avoid once you've named them. If you catch yourself in one, that's normal, and almost every team does early on. Fix the one that's costing you most this week, leave the rest for later, and let your process tighten as you go.
Turn measurement into a system that runs itself
You can run everything above by hand. A spreadsheet, a set of prompts, a weekly hour. For a small program, that's a fine place to start, and starting beats waiting.
The trouble comes with scale. Fifty prompts across five engines, run several times each, every week, benchmarked against competitors, then tied back to the content that closes the gaps. That's a lot of manual work, and it's exactly the part that falls off during a busy month.
This is where a dedicated system helps. DeepSmith runs AI visibility measurement across ChatGPT, Perplexity, Gemini, Claude, and Google AI Mode, reporting mention rate, citation rate, and share of voice with a competitor leaderboard and the sources AI cites most. It surfaces which of your pages actually get cited, and which prompts drive them. Because the same platform also produces on-brand content, the gaps you measure feed straight into what you write next, so measurement stops being a report and becomes a loop. It won't guarantee a ranking or a citation, and no honest tool would claim to. What it does is take the manual weight off tracking, so your time goes to strategy instead of copy-paste.
Not sure where to start? Start small. Run your two-engine audit by hand this week, then decide whether a system earns its place. Keep this AEO measurement guide handy as you go, and work the program one stage at a time: define your metrics, audit, benchmark, monitor, then prove it. You can try DeepSmith free for 7 days and see your real numbers before you commit to anything.
