AI visibility: what it is and how to measure it

// In this article+

AI visibility is whether and how a brand shows up inside an AI-generated answer, not whether a page ranks. Ask ChatGPT, Perplexity, Google's AI Overviews or AI Mode, or Claude a question in your category. AI visibility is the measure of whether your brand gets named, how prominently, what gets said about it, and whether the answer moves the reader toward you. The Interactive Advertising Bureau published a framework for that measurement in August 2026, and its guidelines for measuring AI visibility ↗︎ set the vocabulary the rest of this piece uses.

Most explainers on this topic give away a short manual method, a few dozen prompts run by hand, and then say it does not scale before pointing at a subscription. It scales further than that pitch suggests. This piece works through the IAB's own vocabulary, builds a 20-to-30-prompt set across four engines, and scores it weekly against a scoreboard a marketing lead can run without buying anything.

What is AI visibility?

The IAB defines the measurement, not just the term. Its August 2026 guidelines organize the work around four properties, scored separately rather than blended into one number. Presence asks whether a brand appears at all. Prominence asks how high or how central it appears in a ranked or ordered response. Portrayal asks what is actually said about the brand: accurate, generic, or wrong. Persuasion asks whether the answer moves the reader toward or away from a purchase or a decision, according to the IAB's guidelines for measuring AI visibility ↗︎.

That is a different measurement than SEO visibility ever was. SEO visibility asked whether a URL held a position in a list. One crawl a day could answer that, because the list itself was stable. A generative answer is assembled fresh. It can read differently if the same question is asked five minutes later. Presence is closer to being cited at all than to being ranked, which is also where the naming split happened. The industry calls the discipline of earning that presence by three interchangeable names: generative engine optimization (GEO), answer engine optimization (AEO), and AI search optimization. This piece uses AI search optimization for the rest of it. What actually changes when a page competes for a citation instead of a position is the subject of GEO vs SEO: what changes when the engine answers. This piece stays on the measurement question.

AI ranking, where the phrase gets used at all, usually means something narrower still: a brand's position inside one engine's answer on one run. Presence and prominence, tracked separately across several engines over several runs, is closer to what the IAB means by AI visibility than any single-engine rank ever was.

Why don't AI visibility numbers from different tools agree?

The IAB's own guidelines for measuring AI visibility ↗︎ name the problem plainly: "More than 20 companies now sell AI visibility measurement tools, each using different methodologies". Different vendors run different prompt sets, on different schedules, against different definitions of what counts as a citation. Two dashboards can score the same brand two different ways and both be reporting something real. Both can be right. AI visibility tools sample differently by design, not by mistake, which is why a number from one platform rarely reconciles with a number from another.

That fragmentation is also why almost nobody has this dialed in yet. Only 16% of brands systematically tracked AI visibility when the IAB published its framework, according to PPC Land's reporting on the IAB's framework ↗︎, and the absence of a shared standard is part of the reason. The IAB's response is not to demand one perfect number. It sets up two tiers instead: directional measurement, useful for spotting a trend or flagging a problem, and decision-grade measurement, built to support a budget decision. Without a shared way to evaluate the data, the industry cannot separate a reliable signal from noise, the IAB argues in its announcement of the new measurement standard ↗︎.

That two-tier system makes a manual, unpaid check legitimate. A 20-to-30-prompt set run by hand every week will not produce a decision-grade number. It will produce a directional one, and a directional signal beats the silence most brands are currently sitting on.

How do you measure your own AI visibility this week, without a subscription?

Aleyda Solis updated her AI Search Optimization Checklist ↗︎ on September 20, 2026, with the same prompt-testing discipline the IAB is now standardizing, and it is free to run. Her method: "start with 30 to 50 commercially relevant prompts if you need a minimum viable baseline, grouped by product or service line, audience, market, language and journey stage". Twenty to thirty prompts is enough to start; add the rest once the first round shows which journey stages actually surface your brand.

Build the set by journey stage, not by keyword. Write five to seven prompts for each of three or four stages: someone researching the category with no brand names yet, someone comparing named options, someone asking for a recommendation, and someone asking a specific objection question. A 25-prompt set for a managed IT provider might include "what should a 40-person company look for in a managed service provider," a named comparison between two competitors on a specific requirement, and a direct question about whether your own brand suits a particular buyer.

Run each prompt three to five times per engine, logged out. Cover ChatGPT, Perplexity, Gemini or AI Mode, and Claude. A language model is not deterministic: the same prompt run twice can name different brands. One run tells you what one model said once. Three to five runs, spread across a week, start to show a pattern rather than a coincidence.

Score presence, prominence, and portrayal separately, in a spreadsheet, not one blended average. Presence is the plainest number to start with: how many prompts in the set named the brand at all. Prominence is where in the answer, named first, named in a list, or named only in passing. Portrayal is what was actually said, accurate, generic, or wrong. Aleyda Solis's own five-KPI version of the same idea, from the same checklist, prompt coverage, recommendation rate, linked citation rate, comparative win rate, and representation accuracy, maps closely onto the IAB's four properties and is worth adopting as the column headers. Tracking prompts by hand this way is also the discipline behind Search Engine Land's list of eight GEO metrics ↗︎, which groups the same signals under AI citation frequency, share of model voice, prompt coverage, and answer inclusion rate.

A small operator can move this number. A September 2026 paper out of Northwestern on brand retrieval ↗︎ built two new metrics, BRP@k and MRR@k, to measure brand retrieval and ranking across large language models. The study covered six models and five product categories, through repeated sampling. Its finding: "Prominence is associated with broader marketplace-visibility signals, particularly search interest and online brand conversation", not with incumbent brand size. A smaller brand with active search interest and real conversation around it has a documented path to prominence. That path has nothing to do with ad spend or headcount. What actually decides whether a model names a source once a passage is retrieved is a separate, mechanical question, covered in how a model decides what to cite.

What do the early benchmarks actually mean?

A low first-week score is not evidence the method failed. Citation presence in US ChatGPT prompts rose from about 1.6% in June 2025 to roughly 6.8% by May 2026, tracked by Similarweb's own stated method ↗︎. That is a fourfold increase over eleven months, and it is still a small share of prompts. A brand appearing in a handful of the twenty to thirty prompts in its first weekly run is closer to that baseline than to a failure.

AI answer engines are also far more selective than Google ever was. A report on local visibility ↗︎ found only 1.2% of locations recommended by ChatGPT, 11% by Gemini, and 7.4% by Perplexity, against 35.9% of the same brands appearing in Google's local 3-pack. Google ranks broadly and lets the searcher scroll. An AI engine picks a handful of names and stops there. A citation rate that looks disappointing next to an old SEO benchmark is often normal next to the current AI one.

The referral traffic that used to confirm visibility indirectly is also getting harder to read. Google is testing anchor links inside AI Overviews ↗︎ that route into AI Mode rather than out to a webpage, keeping the follow-up question inside Google's own interface. That test weakens another click-based proxy for visibility. Tracking presence through prompts, rather than waiting for a referral to show up in analytics, is the more durable method precisely because the click is disappearing from the path.

What to do this week

1. Write 20 to 30 prompts grouped by journey stage, category research, named comparison, recommendation, and objection, following Aleyda Solis's grouping method ↗︎.
2. Run each prompt three to five times across ChatGPT, Perplexity, Gemini or AI Mode, and Claude, logged out or in an incognito session, so personalization does not skew the result.
3. Score presence, meaning how many prompts named the brand, prominence, and portrayal in separate columns rather than one blended number, using the IAB's four properties ↗︎ as the scoring vocabulary.
4. Treat the first month of results as directional, not decision-grade, per the IAB's own tier system: a trend worth watching, not a number to defend in a budget meeting.
5. Re-run the set weekly. Citation share moves more like Similarweb's fourfold rise in ChatGPT citation presence ↗︎ than like a static ranking, and Google's test of AI Overview links that open AI Mode ↗︎ is a reminder that the surfaces themselves keep moving too.
6. Hand the spreadsheet, the prompt set, and the weekly cadence to an AI SEO agency once the volume of prompts, engines, and product lines outgrows a person doing it by hand.

Run this way, the routine is the simplest form of AI visibility monitoring: no dashboard, just a prompt set, a spreadsheet, and a repeat schedule. Search Engine Land's argument for visibility systems ↗︎ fits at this scale too: treat the work as a running organizational discipline, not a one-time audit.

FAQ

Is AI visibility the same thing as GEO?

Not exactly. AI visibility is the measurement, whether and how a brand appears inside an AI-generated answer. GEO, AEO, and AI search optimization describe the practice of earning that appearance. One is the scoreboard, the other is the game.

How often should I re-run the prompt check?

Weekly, at minimum, since citation share moves with model updates rather than on a fixed schedule. Treat the first month of results as directional rather than decision-grade, in line with the IAB's own two-tier framework, before drawing a firm conclusion from any single week's numbers.

Do I actually need a paid AI visibility tool?

Not to start. A 20-to-30-prompt set run by hand across four engines produces a real, if directional, signal, and only 16% of brands were running even that much when the IAB published its framework. The manual method gets harder to sustain once the prompt set grows past a few dozen, spans multiple languages or markets, or needs to run daily instead of weekly, which is usually the point a paid tool earns its cost.