How to Measure AI Visibility: The Honest Guide to Every Method
Turn this article into takeaways for your work.
Each assistant summarizes the article only for you and suggests best practices for your work.
To measure AI visibility, run a fixed set of real buyer questions against ChatGPT, Gemini, Perplexity, and Copilot on a schedule and log whether you're mentioned, or pay a monitoring tool to run that sampling at scale. Check server logs for AI crawler activity to confirm pages are at least reachable. Then accept the limit up front: nobody, including the vendors selling a dashboard, can yet tell you whether a citation drove revenue, what your true share of AI answers is, or whether a change you made caused a change you saw.
This category is younger than it looks. There is no first-party console from any AI vendor that works the way Google Search Console works for organic search, so every method below is a workaround, not a finished instrument. That's worth saying plainly before you budget around any of them.
The Four Methods, at a Glance
| Method | Cost | What it actually shows | Where it misleads |
|---|---|---|---|
| Manual prompting | Free (your time) | Whether you're mentioned for the specific prompts you ran, on the day you ran them | A sample of one session, not a census. Answers vary by account, location, and the moment you ask |
| Paid monitoring tools | $20 to $780+/month, or custom | Sampled citation tracking across a defined prompt set, repeated on a schedule | Each vendor samples differently, so two tools can honestly disagree about the same brand |
| Server logs and analytics | Free, if you already have log access | Confirms a page was fetched and reachable; referral sessions that do click through | A fetch is not a citation, and most AI-driven visits never show up as a referral at all |
| Google Search Console | Free | Organic impressions, clicks, and position for Google's "Web" search type, which already includes AI Overviews traffic mixed in | No way to isolate what came from an AI Overview versus a classic blue link, and zero visibility into ChatGPT, Perplexity, or Copilot |
Run more than one, and run it across more than one assistant. A synthesis of roughly 680 million tracked citations found only about 11% of the domains ChatGPT cites also show up in Perplexity's citations (5W AI Platform Citation Source Index 2026), so measuring your standing in one assistant tells you almost nothing about where you stand in another. Each method below also covers a gap the others leave open, and none of them, even combined, closes every gap.
Method 1: Manual Prompting
This is the method every paid tool automates, so it's worth understanding before paying anyone to do it faster. Pick 10 to 20 real buyer questions, ask them across ChatGPT, Gemini, Perplexity, and Copilot, and write down what comes back. It costs nothing but time, and it's the only method where you see the actual answer text, not a vendor's summary of it.
A starter prompt set. Swap in your own category, brand name, and competitors.
| Category | Example prompt | What it tests |
|---|---|---|
| Category roundup | "What are the best [category] tools for a [company size] company?" | Whether you surface in an unprompted best-of list |
| Problem-first | "How do I solve [specific pain point] for my team?" | Whether you show up as a solution before anyone names a brand |
| Direct lookup | "What is [Brand Name] and is it any good?" | Whether the facts the assistant states about you are accurate |
| Head to head | "[Brand Name] vs [Competitor]: which should I pick?" | How you're framed next to a named rival |
| Pricing | "How much does [Brand Name] cost?" | Whether the stated price is current |
| Alternatives | "What are alternatives to [Competitor]?" | Whether you appear as a substitute for a rival's buyers |
| Use case fit | "Best tool for [specific job] at a [industry] company" | Visibility in a narrower, more buyer-realistic query |
| Review style | "Is [Brand Name] worth it? What do people say about it?" | Whether sentiment and review content gets surfaced |
| Open recommendation | "I'm a [persona] trying to [outcome]. What should I use?" | The most natural phrasing, with no brand named by you at all |
| Correction check | "Is [Brand Name] still [outdated fact, e.g. a feature you've since changed]?" | Whether stale information about you is still circulating |
A recording template. One row per prompt, per assistant, per run.
| Date | Assistant | Prompt category | Mentioned? | Position in answer | Competitors named | Direct link or citation? | Notes |
|---|---|---|---|---|---|---|---|
| 2026-10-02 | ChatGPT | Category roundup | Yes | 2nd of 5 | Competitor A, Competitor B | No | Named but no link; pricing stated was wrong |
| 2026-10-02 | Perplexity | Pricing | No | Not mentioned | Competitor A | N/A | Only one competitor surfaced at all |
Repeat monthly at minimum. A single run tells you almost nothing, since the same prompt can return a different answer an hour later depending on retrieval state, account history, and location. The trend across several runs is the actual signal, not any one session.
Method 2: Paid Monitoring Tools
A paid AEO tool is the same manual prompting exercise run at scale: a larger prompt library, a repeating schedule, a dashboard on top. That's the entire product. It earns its price once you need daily or weekly checks across dozens of prompts rather than a handful run occasionally.
| Vendor | Entry price | What it covers |
|---|---|---|
| Rankscale | $20/mo, or $17/mo on yearly billing (Essentials) | Credit based, 17+ engines claimed; no engine add-ons, but credits are weighted per engine (0.25 for ChatGPT, Perplexity and Gemini, 2 for Claude) |
| Otterly.ai | $29/mo (Lite) | 15 prompts, 4 core engines (ChatGPT, AI Overviews, Perplexity, Copilot); extra engines sold separately from $9/mo |
| HubSpot AEO | $50/mo ($45/mo annual) | Beta; 25 prompts across 3 engines (ChatGPT, Gemini, Perplexity) and no Google AI Overviews; the free trial also tracks 25 prompts, no card |
| Ahrefs Brand Radar | From $50/mo (Custom Prompts add-on), free on paid Ahrefs plans; currency is region-served (USD, EUR and GBP have all appeared) | Prompt allowance scales with the Ahrefs plan you're already on (5, 10 or 20 prompts on Lite, Standard, Advanced); covers ChatGPT, Gemini, Perplexity, Copilot, AI Overviews, AI Mode and Claude (8 checks per update), with Grok not named |
| Trakkr | $100/mo (Growth) | 1 brand, 50 daily prompts (more in paid packs), all 8 named AI models included, no per-model fee |
| Semrush AI Visibility Toolkit | $99/mo per domain on annual billing (its knowledge base says $99 a month) | Semrush is now an Adobe company (acquisition completed 28 April 2026). 25 tracked prompts across 4 engines, folded into the Semrush dashboard; 50 more prompts cost $60/mo per its knowledge base |
| AthenaHQ | Free trial credit, then $295/mo | Essential tier gives a one-time $25 credit; Starter's 3,600 monthly credits cover 11+ platforms |
| Peec AI | $95 / $245 / $495 | Prices are public but render in JavaScript: 50 / 150 / 350 prompts, unlimited users on every tier, 3 models per tier from ChatGPT, AI Mode, AI Overviews, Copilot, Gemini and Naver AI; Perplexity costs extra and Claude is Enterprise-only |
| Conductor | Not published | An SEO platform with AEO added; three tiers, and every one routes to a demo or sales |
| Goodie | Core $399/mo, Pro $999/mo, agency Growth $275/mo, all on annual billing | Monthly or quarterly billing is $478.80, $1,198.80 and $330. Core is self-serve with a 7-day trial; Pro and Enterprise are demo-gated |
| Profound | No base price (trial free, Enterprise custom); agency add-ons are listed, including client workspaces at $399/mo | Repositioned in 2026 from a tracking dashboard to an "AI Marketer" agent; no self-serve paid brand tier exists at any volume |
Full vendor profiles, per-tool strengths and limits, and a real-usage cost scenario are in our AEO tools comparison.
What you're buying is scheduled prompting at scale plus a dashboard, not a ground truth. Two tools checking the same brand on the same day can honestly disagree, since each samples a different prompt library against a different account and location. If a vendor's "visibility score" lands in a board slide, say plainly that it's that vendor's formula, not an industry standard. Nobody has agreed on how to compute one, and the numbers aren't comparable across tools. Which tools verify a linked citation and which only count your name is the main split, covered in Best LLM Citation Tracking Tools in 2026, and Best ChatGPT Rank Tracking Tools in 2026 explains how each tool samples ChatGPT.
Method 3: Your Own Server Logs and Analytics
The most underused method of the four, and free if you already have log access. It answers a narrower question than the others: not whether you were cited, but whether your pages were even reachable by the systems doing the citing.
What to search your logs for.
| Operator | User agent | What a hit tells you |
|---|---|---|
| OpenAI | GPTBot | Training-data crawl. Not used to decide ChatGPT search inclusion |
| OpenAI | OAI-SearchBot | Indexing crawl for ChatGPT's own search feature |
| OpenAI | ChatGPT-User | A live, user-triggered fetch, not a scheduled crawl |
| Anthropic | ClaudeBot | Indexing crawl. A 12-week study across 83 sites found it fetched robots.txt 3,120 times and llms.txt only 9 times on the same sites |
| Perplexity | PerplexityBot | Indexing crawl for Perplexity's own search |
| Perplexity | Perplexity-User | Live, user-triggered fetch; generally ignores robots.txt since a person asked directly |
| Googlebot | The single crawl feeding Search, AI Overviews, and AI Mode together, with no separate AI-only crawler to isolate | |
| Microsoft | Bingbot | Crawls and indexes for Bing; Copilot's answers run on that same index |
Crawler names and the robots.txt versus llms.txt fetch counts come from ezy.ai's 12-week, 83-site study, published July 2026. If you've also published an llms.txt file, don't expect many hits against it either way: a May 2026 Ahrefs analysis of 137,000 domains found 97% of published llms.txt files received zero requests at all.
A crawler hit only proves a page was available to be pulled, not that it was pulled into a cited answer. Logs confirm reachability and nothing past it. Request logs are also how researchers found that AI crawlers almost never fetch llms.txt; llms.txt Explained has the numbers.
Referral traffic understates itself, and being cited rarely means being clicked. GA4 or server logs can isolate sessions where the referrer domain is chatgpt.com, perplexity.ai, or similar. That traffic is real and converts well, but it's a floor, not a full count. Similarweb's 2026 tracking found AI platforms driving 770.7 million referral visits per month worldwide, up 117.4% year over year, and its own report states plainly that "some AI visits arrive without referrer data and land in direct traffic, so referrer-based counts understate true volume" (Similarweb). Adobe Analytics, reviewing more than a trillion US retail visits, found AI-referred traffic to retail sites up 138% year over year by May 2026, converting 54% better than non-AI traffic (Digital Commerce 360). Even when a citation does appear, most readers never act on it: Pew Research Center tracked 900 US adults across roughly 69,000 Google searches and found a standard result got clicked just 8% of the time when an AI summary appeared, versus 15% when it didn't, and only 1% of visits included a click on a source cited inside the summary itself. The channel is real and converts well. The number your analytics shows you is still smaller than what's happening.
There's a bigger reason your logs only see part of the picture: most of what gets cited in a category reportedly doesn't live on the brand's own domain at all. Noble, which sells citation placement, claims on its homepage that 85% of AI citations are third party, with no linked study or sample size. DerivateX, an SEO and GEO agency, published a more rigorous 88.4% from a disclosed study of 233 recommendations across 40 B2B SaaS categories. Both figures come from companies selling the fix they describe, and no neutral measurement exists; treat the direction as plausible, the percentage as unsettled. Either way, if most citations sit on third-party pages, your logs were never going to see the majority of your AI visibility. Our breakdown of paying for AI citations covers that claim in full.
Method 4: Google Search Console, and Its Real Limit
Search Console is free and already installed at most companies, but it covers less of this problem than its interface suggests. Google's own documentation is direct about the limit: pages appearing in AI features, including AI Overviews and AI Mode, are folded into regular organic traffic and reported inside the Performance report's existing "Web" search type, not a separate AI view (Google Search Central, "AI features and your website"). There's no filter to isolate "clicks that came from an AI Overview." Google's own guidance for that level of detail is to track conversions and time on site in a separate analytics tool instead.
| What Search Console shows | What it doesn't show |
|---|---|
| Aggregate organic impressions, clicks, and average position, including traffic from AI Overviews mixed in | A separate row or filter for AI Overview or AI Mode appearances specifically |
| Index coverage and crawl issues for Googlebot | Any data at all for ChatGPT, Perplexity, Gemini's standalone app, or Copilot, which Google has no visibility into |
| Query-level and page-level performance trends over time | Whether a given impression or click was generated by a classic result or an AI-generated summary |
Search Console remains a real, free signal for whether Google can find and rank your pages at all, which matters because AI Overviews draw from the same index Search already uses. It is not, and currently can't be, a dedicated AI visibility tool. Our SEO metrics guide covers the rest of that Performance report, still the foundation this sits on top of.
What You Genuinely Cannot Measure Today
Say this out loud in any planning meeting before someone budgets against a number that doesn't exist yet.
| You want to know | Why no method above answers it |
|---|---|
| Did this specific citation generate revenue | No assistant publishes a click log tied to a conversion event, and the referral traffic that does arrive is itself undercounted |
| What our true share of AI answers is in our category | Every method here samples a prompt set. Nobody, including the vendors, has a complete record of every question real buyers actually asked |
| Whether a change we made caused a citation change we saw | Model versions and indexes refresh on their own schedule, not yours. A shift in the same month as your content update could be the update, the refresh, or both |
| How much of our visibility comes from mentions we don't control | Most citations in a category reportedly point to third-party pages rather than the brand's own site, and no tool here can fully see a mention on someone else's page unless its sampled prompts happen to surface it |
If a vendor's pitch skips past this table entirely, that's worth noticing on its own.
What a Defensible Report to Your CEO Actually Looks Like
A credible update names its method and its limits in the same breath as the result: "We ran 15 buyer-intent prompts across ChatGPT, Gemini, and Perplexity this month. We were mentioned in 6 of 15, up from 4 last month. [Competitor] appeared most often. This is a sampled check against a fixed prompt list, not a full measure of every conversation about us, and we can't yet tie a mention to a specific deal." That beats a single "visibility score," because it tells the room how much weight the number can bear.
Two habits keep it honest over time: never report a vendor's proprietary score as an industry standard, say whose formula it is, and keep the method identical period to period, same prompts, same assistants, same cadence, since changing the sample makes month-over-month comparisons meaningless. A tool like the ones in our AI brand monitoring roundup helps keep that cadence consistent without someone remembering to run the prompts by hand.
Frequently Asked Questions about Measuring AI Visibility
What's the single cheapest way to start measuring AI visibility?
Manual prompting. Pick 10 to 20 real buyer questions, run them across ChatGPT, Gemini, Perplexity, and Copilot, and log whether you're mentioned. It costs nothing but time and is the same exercise every paid tool automates at scale.
Can Google Search Console show me my AI Overview performance separately from regular search?
No. Google's own documentation says pages appearing in AI features, including AI Overviews and AI Mode, are folded into the existing "Web" search type in the Performance report, not broken into their own view. Search Console also has zero visibility into ChatGPT, Perplexity, or Copilot.
Why do two different AEO monitoring tools report different numbers for the same brand?
Each vendor samples a different prompt library against different accounts and locations on its own schedule. That's a sampling difference, not necessarily an error, which is why a proprietary "visibility score" from one tool isn't comparable to another's.
Does seeing GPTBot or PerplexityBot in my server logs mean I got cited?
No. A crawler hit confirms your page was fetched and available to be used, not that it was pulled into a cited answer. Treat log activity as a reachability check, not a citation count.
Can I measure whether an AI citation actually drove revenue?
Not reliably yet. There's no click-pixel or conversion log tied to an AI assistant's retrieval step. Referral traffic that does click through shows up in GA4 or server logs, but industry trackers describe that number as understated, since many AI-driven visits arrive without referrer data at all.
Is it true that most AI citations go to third-party sites instead of a brand's own domain?
It's widely repeated, most prominently an unsourced 85% figure from Noble, which sells citation placement, and a more rigorous but still interested 88.4% figure from the agency DerivateX. Both sell the fix for the gap they describe. No neutral measurement exists, though the direction is plausible and explains why your own analytics only ever see part of your AI visibility.
What to Do Next
Start free before you pay anyone (our roundup of free AI visibility tools covers what's genuinely free). Run the manual prompt set above for a month, check your logs for the crawler user agents listed, and pull your Search Console data with the limit in mind. If that month shows you're genuinely absent and you need the check running daily instead, our AEO tools comparison has vendor-verified pricing for the next step up. If the worry is accuracy rather than presence, Best AI Brand Monitoring Tools in 2026 tests for it. Still sorting out what AEO even changes versus classic SEO? Read AEO vs SEO and GEO vs AEO vs SEO first, and our guide to getting cited in AI answers covers tactics once you know where you stand. Whatever you pick, report the method next to the number every time. A visibility figure with no stated sample size is a marketing claim wearing a metric's clothes.
