← All guides

How do you measure AIO?

Updated September 25, 2026 · by Abdul Jaafar

You measure AIO with three separate numbers, because no single one survives contact with how AI search actually works: a readiness score for whether AI systems can reach and quote your pages, a mention rate for how often they actually name you, and a presence score for the off-site conditions that decide the gap between the two. The first is free and takes about fifteen seconds. The second takes an afternoon of running real queries and writing down what comes back. Most tools in this category sell you one of the three and call it AI visibility.

Prefer to watch? The 2:48 walkthrough at the end of this guide runs both steps on screen.

Why can’t one number measure AIO?

Because readiness and outcome are different things, and a site can max one while failing the other completely. Readiness lives on your own pages: can a crawler fetch you, can a model lift a clean passage out of you. Outcome lives in the answer: does the assistant actually say your name when someone asks.

A site can score 97 on readiness and be named in zero answers, because the remaining decisions get made off your site entirely. It can also happen in reverse: a business with a mediocre site gets named constantly because it sits on the directories, roundups, and review pages that AI already trusts for that category. Collapse those two things into one number and it has to lie about one of them.

Scale matters here too. When we scanned the web’s top 1,500 domains, the median AIO Score was 72 and 45% of sites graded D or F, with the full breakdown in the study. Readiness is where most sites are quietly losing, which is exactly why it is the cheapest number to fix and the wrong one to stop at.

Which three numbers should you track?

AIO Score is readiness, 0 to 100, from 21 deterministic checks covering whether AI can reach your pages and whether it can quote them. Same site, same score, twice. This is the one you can get free and instantly.

AI Mention Rate is the observed outcome: the share of live AI answers in which you are named, broken down per engine and per topic. It reads like “named in 3 of 24 answers, ChatGPT 0 of 6, Perplexity 2 of 6.” Nothing estimates this. You either run the queries or you don’t know.

AI Presence Score covers the off-site gates that explain the distance between the first two: whether you rank for the sub-queries AI generates, whether it recognizes you as an entity, whether you appear on the sources it already cites, and what your reputation looks like across platforms. Those four are weighted 25, 20, 35, and 20, with off-site references carrying the most because they are the most repeated lever in the evidence.

How do you measure readiness?

Run a scan and read the checks, not just the number. Our free scan returns all 21 results in about fifteen seconds, and the methodology publishes every check, every point weight, and the research behind each one, so you can re-derive your own score by hand if you want to.

That re-derivation test is the whole point. A readiness score you cannot audit is worse than no score, because it directs your effort by authority instead of evidence, and this category has a real fake-score problem. If you are choosing between tools, what an AI score actually is sorts the two families apart and gives you the test to apply.

How do you measure whether AI actually names you?

By hand, with repeated live queries, which sounds primitive and is currently the only honest method. The procedure:

1. Pick four to six topics a buyer would really ask about. Your category in your city, best-of lists, the problem you solve phrased as a complaint, cost, comparison against a named competitor, and brand due diligence.
2. Ask each one several ways, on each engine. ChatGPT, Perplexity, Gemini, Claude, and Google’s AI Mode all retrieve differently, and a result on one says nothing about the others.
3. Repeat the runs. AI answers are non-deterministic and personalized, so a single prompt tells you almost nothing.
4. Record four things per answer. Whether you were named, where you sat in the list, whether you were cited with a link or only mentioned, and which sources the answer drew on.
5. Divide named by total. That is your mention rate. The sources you logged become your citation map: the exact pages AI uses to answer your category, which you can then check one by one for whether you appear on them.

Report the run count alongside the number, because this is a sampled estimate and not a census. If you would rather not spend the afternoon, the AI Presence report is this process done for you across five engines with a person reviewing every answer.

What shouldn’t you measure?

The three metrics that generate the most invoices and the least signal.

llms.txt requests, because almost nothing reads the file. Ahrefs found 97% of llms.txt files received zero requests across a month of logs on 137,000 domains, and Otterly’s 90-day experiment counted 84 hits out of 62,100 AI bot visits.

Schema coverage, because LLM parsers strip it to plain text. A peer-reviewed controlled study found it moved nothing on Google or Maps, with one modest ChatGPT-only effect. Both of these are fine to have and neither is a measurement.

The single-prompt screenshot. Asking ChatGPT once, getting named, and declaring the problem solved is a sample size of one against a system that answers the same question differently on Tuesday. It is the most common way businesses convince themselves they are visible.

How often should you re-measure?

Readiness after every fix, mention rate quarterly. Readiness is free, instant, and deterministic, so rescan the moment you change a robots.txt rule or rewrite an opening paragraph and watch the number move. Mention rate moves slowly and carries sampling noise, so measuring it monthly mostly produces the illusion of change. Quarterly is enough unless you are actively working the off-site gates, in which case monthly runs on the same fixed topic set will show you real movement.

Start with readiness, because it is the floor. If AI cannot fetch or quote your pages, no amount of off-site work will fix your mention rate, and you will have spent a quarter measuring the wrong end of the problem. If you want the vocabulary first, what AIO means in SEO covers the practice this page measures.

2:48