AI search visibility is how often, how prominently and how accurately your brand appears in answers from ChatGPT, Gemini, Claude, Perplexity, Grok and Google’s AI Overviews and AI Mode. Measuring it is harder than measuring classic SEO. There is no rank to look up, most engines offer site owners no reporting at all, and the same question can get a different answer every time someone asks it.
That does not make it unmeasurable. It means you need a defined sample, a small set of KPIs that each answer a different question, and a cadence that separates signal from noise. This guide shows how to measure AI search visibility with all three, and covers the measurement mistakes that most often lead teams to the wrong conclusion.
Key takeaways
- Measure AI search visibility by sampling: a fixed set of prompts, run on a schedule, across the engines your buyers use.
- Use a layered set of KPIs. Mentions, citations, position, sentiment and share of voice describe the answers. AI referral visits and conversions describe the business result.
- No single KPI tells the full story. A brand can be mentioned and never cited, or cited often and described badly.
- Learn your normal noise before you report a change, and always compare like with like.
- Pair a light daily alert layer with weekly reviews, monthly scorecards and quarterly resets, each with a named owner.
Why AI search needs its own measurement
Classic SEO reporting rests on rankings, impressions and clicks. Those still matter, because many AI systems retrieve from web indexes, but they do not capture what happens inside an AI answer. Three differences shape how you measure.
Answers vary from run to run
Generative engines write new text each time. The brands named, their order and the sources cited can all change between runs, between engines and between locations. One screenshot is an anecdote. You need repeated runs over time to see a pattern, and a sense of how much the numbers move when nothing has changed.
Visibility does not always produce a click
An assistant can recommend your brand by name without linking to you. The reader may remember you and search later, or may never visit at all. If you only measure traffic, you miss the influence AI answers have on which brands get considered in the first place.
Most engines do not report to you
Google’s Search Console now has a generative AI performance report that shows impressions in AI Overviews and AI Mode, but most assistants give site owners nothing comparable. Analytics tools see only the visits that arrive with a recognizable referrer. That makes your own tracked sample of answers the main source of truth for what engines say about you.
The core KPIs for AI search visibility
These seven KPIs cover what the answers say about you and what that does for the business. Define each one once, write the definition into your report and keep it stable.
1. Mentions and mention rate
A mention is an answer that names your brand. Mention rate is the share of answers in your sample that name you. It is the broadest measure of whether engines know you and consider you relevant to the question.
Track it on unbranded prompts. Prompts that include your name will almost always mention you, which inflates the number without telling you anything about how buyers discover you.
2. Citations and citation rate
A citation is a link to one of your pages that the engine includes as a source. Citation rate is the share of answers that cite at least one of your URLs. Citations show which of your pages engines trust as evidence, and they are the main path from an answer to your site.
Track cited URLs, not just counts. Knowing that your pricing page is cited often and your comparison page never is far more useful than a total. See brand mentions vs citations for how the two signals differ and why you need both.
3. Position
When an answer names several brands, position records where you appear. Being named first is not the same as being named sixth. Track your average position in the answers that name you, and how often you appear first. A brand that is named often but rarely early has a different problem from one that is rarely named at all.
4. Sentiment and accuracy
Sentiment labels each mention as positive, neutral or negative, and net sentiment summarizes the period. Read it together with the actual text around your name, because automated labels can misread nuance and a short snippet can hide context.
Accuracy deserves its own check. Is your pricing current? Are your features, integrations and availability described correctly? Wrong claims are worth fixing at their source even when every other number looks healthy, because they shape decisions quietly.
5. Share of voice
Share of voice compares your mentions with those of the competitors you track. It puts everything else in context: your mention rate can rise while your share of voice falls, which means a competitor is rising faster than you. Our guide to AI share of voice covers formulas and how to set benchmarks from your own data.
6. AI referral visits
AI referral visits are the sessions that arrive on your site from an AI assistant. Track them by assistant and by landing page. Pages that are cited but receive no AI visits deserve a second look, and so do pages that receive AI visits you did not expect.
Referrer data undercounts. Some assistants and apps pass no referrer, people copy links, and some readers search for your brand after seeing it in an answer. Treat the number you can detect as a minimum. How to track AI traffic in GA4 explains the setup and its limits.
7. Conversions from AI visits
Conversions close the loop: signups, demo requests, purchases or other key events from visitors who arrived through an assistant. A small number of AI visits that convert well can matter more than a large number that leave immediately. Compare AI conversion rates with other channels using exactly the same definitions.
Tip: Many teams roll mention rate, citation rate and position into one composite score for reporting. That works well as a headline, but keep the parts visible. When the score moves, the parts tell you why.
Supporting signals worth watching
Beyond the core KPIs, a few supporting signals help you explain what you see. None of them is a visibility metric on its own, but each adds context.
- AI crawler activity: which AI bots fetch which pages, taken from your server logs. If an engine’s crawler never reaches a page, that engine is unlikely to cite it. Keep crawler hits out of visitor counts.
- Search Console data: Google clicks, plus impressions in AI Overviews and AI Mode from the generative AI performance report. Compare these trends with your AI referral visits.
- Branded search demand: a sustained rise in searches for your brand after a period of strong AI visibility is consistent with AI influence. It is supporting evidence, not proof.
- Self-reported attribution: a “how did you hear about us” field with an AI assistant option catches visits that left no referrer.
- Competitor citations: the sources cited when competitors are named, which point to gaps you can close.
Choose the KPIs that match your goal
All seven KPIs are worth tracking, but your goal decides which one leads your reports. Picking a lead KPI keeps reviews focused and stops a good number in one area from hiding a weak one that matters more.
If you want to be considered more often
Lead with mention rate and share of voice on unbranded, recommendation-style prompts. Watch position closely: moving from fifth to second in the answers that already name you can matter as much as appearing in new ones. This fits brands in crowded categories, where the shortlist decides who gets a demo or a visit.
If you want more qualified traffic
Lead with citation rate and AI referral visits by landing page. The question is which pages engines cite and whether those citations turn into clicks. Pages that are cited often but send few visits may need a clearer title, a stronger opening or a more specific promise.
If you want revenue you can attribute
Lead with conversions from AI visits, then work backward. Find the assistants and landing pages that produce signups or orders, and check which prompts and citations feed them. Keep the volume caveat in mind: AI traffic is still small for many sites, so judge conversion rates over months rather than weeks.
If you want to protect your reputation
Lead with sentiment and accuracy. Review the text around your mentions every week, log every wrong claim with its likely source and track how many you have corrected. This matters most after a pricing change, a rebrand or a product launch, when engines may keep repeating old facts.
If you report to clients or leadership
Lead with one composite score and share of voice, because both are easy to read at a glance, and keep the underlying KPIs one click away. Explain each movement with the parts that caused it.
How to set up measurement you can trust
The KPIs are only as reliable as the sample behind them. Spend time on the setup, because changing it later breaks your trend lines.
Choose a fixed prompt set
Start with 20 to 50 unbranded prompts that real buyers ask, grouped into four to seven topics. Cover the buying journey: problem questions, best-option questions, comparisons and pricing. Keep the set stable once you start, add new prompts deliberately and note the date when you do. Our guide to tracking brand visibility in ChatGPT walks through building the set.
Track the engines your audience uses
For most brands that means ChatGPT and Google’s AI Overviews and AI Mode, followed by Gemini, Perplexity, Claude and Grok. Track each engine separately. Each has its own sources and habits, and a rollup hides where you are weak.
Include location and language where it matters
Answers differ by country and by language. If you sell in several markets, track prompts in each one, written in the language buyers actually use, rather than translating one English set word for word.
Set a baseline and learn your noise
Run the full set daily for two to four weeks before you change anything. Record each KPI per engine and per topic. Note how much the numbers move week to week with no intervention. That spread is your noise floor, and changes smaller than it are not results.
Decide how you count
Write down the counting rules before the first report. Count a brand once per answer, even if the answer names it several times. Decide whether a mention inside a citation title counts as a mention. Decide whether you report each engine separately, an average of engines or both. Small rules like these change the numbers, and they are easy to forget when someone new joins the team.
Compare like with like
When you report a change, compare the same prompts, engines and regions across both periods. If prompts were added or paused, leave them out of the comparison. Otherwise a change in the sample looks like a change in visibility.
Build a reporting cadence
Measurement only creates value when someone acts on it. A layered cadence keeps the effort low and the response fast, and gives each audience the level of detail it needs.
Daily: alerts, not dashboards
You do not need to read a dashboard every morning. Set alerts for the events that need a quick response: a sharp drop in visibility, a lost citation on an important page, a competitor surge or a competitor moving above you. Everything else can wait for the weekly review.
Weekly: a short working review
Spend 20 to 30 minutes on trends with the people who do the work. Look at movement per engine and topic, new competitor gains, citations won and lost, and AI referral visits by landing page. End with the three most important things to fix, each with an owner and a date.
Monthly: a stakeholder scorecard
Summarize the month for people outside the working team. Show each core KPI against the baseline and the previous month, the wins, the losses and what you plan next. Produce it on the same day each month so comparisons stay consistent. Agencies can use the same structure for each client, as the agency client reporting playbook describes.
Quarterly: reset and recalibrate
Once a quarter, review the prompt set, the competitor list and your targets. Add prompts for new products or markets, retire ones that no longer reflect how buyers ask, and check whether new competitors deserve a place. Record what changed, so the next quarter’s comparisons stay honest.
Tip: Connect every reported change to an action. “Share of voice in the pricing topic fell after a competitor’s new comparison page started being cited in Perplexity, and we are publishing a response” is a report people can use. “Share of voice fell three points” is not.
Common measurement mistakes
- Judging from one screenshot. A single answer cannot represent an engine, let alone all of them.
- Tracking branded prompts only. They measure people who already know you, not discovery.
- Changing the prompt set constantly. Every change breaks the trend line.
- Treating one engine as representative. Visibility in ChatGPT says little about Gemini or AI Mode.
- Reporting a composite number without its parts. A score can hold steady while mentions rise and citations fall.
- Ignoring noise. Small week-to-week moves are usually normal variation, not wins or losses.
- Counting only referral traffic. Mentions without links still shape decisions, and many AI visits arrive with no referrer.
- Mixing bots with people. AI crawler visits are not visitors. Keep them out of traffic and conversion reports.
- Claiming causation too quickly. If visibility rises after a page update, check whether similar prompts you did not touch moved too before you credit the change.
- Collecting data nobody acts on. Every weak area needs an owner and a next step.
How aeoora helps
aeoora checks the AI engines daily for your tracked prompts and turns the answers into the KPIs above, with each one broken down by engine, topic, prompt and location.
- AI Visibility Score: one number from 0 to 100 built from mention rate (60 percent), citation rate (25 percent) and position (15 percent), shown for your brand, each prompt, each topic and each competitor. Read how the score works.
- Why it changed: splits a change between two periods by engine, prompt and topic, comparing only prompts answered in both, so adding or pausing prompts does not look like a rise or a fall.
- Mentions, sentiment and citations: mention rate, position, net sentiment with labels you can correct, cited domains and URLs, and competitor gaps.
- AI referral traffic and conversions: a tracking snippet counts visits from each assistant without cookies by default, and optional conversion tracking credits signups and orders to the assistant that first sent the visitor. Google Analytics and Search Console connections add key events, revenue and Google clicks.
- The cadence, built in: the Daily Pulse and Slack alerts cover the daily layer, and Reports produce fixed snapshots and PDFs for weekly and monthly reviews.
- Evidence of impact: the “What aeoora did for you” report compares each shipped change on its target prompts before and after, and against similar prompts nothing touched.
Next steps
- Write your KPI definitions on one page and share it with everyone who reads the reports.
- Build or review your prompt set, then run a two to four week baseline.
- Set up AI referral tracking so visits and conversions sit next to visibility.
- Put the daily, weekly, monthly and quarterly rhythm on the calendar, with owners.
Useful docs for the setup: visibility reports, Daily Pulse and alerts, AI referral traffic and conversions from AI visits. The free AI referrer checker helps you identify assistant visits in your own analytics. Compare plans on the pricing page, or start a 2-day free trial to see your own baseline.


