Google AI Overview
What It Does
Shows whether Google's AI Overview — the generated answer above the blue links — names your business for the searches you chose to watch, and which sites it drew on instead when it did not. This is NOT the same thing as the Visibility Audit, and the difference matters more than the similarity. • THE VISIBILITY AUDIT asks the engines directly, through their APIs. Repeatable, and it can carry a confidence range. • THIS reads a live Google results page. Different retrieval, different sources, different behaviour. Both are worth knowing. Averaging them would give you a number that describes neither, so the product refuses to combine them — a trend will not join readings from different sources, and there is no blended "AI score" quietly containing both.
Before you start: it runs on your own DataForSEO account
These checks are bring-your-own-key, and that is deliberate — it is how the plan price stays where it is instead of carrying everyone's lookup costs. 1. Create an account at dataforseo.com and add credit. Their minimum deposit is $50. 2. In Settings, enter your DataForSEO account email and your API PASSWORD. 3. Use the API password from their dashboard — NOT the base64 string in their welcome email. That string is your email and password already combined, and pasting it will simply fail to authenticate. Cost is about $0.0012 per search checked. Your plan carries a monthly budget for these — $10 on Starter, $40 on Growth and Agency, and whatever you set on Enterprise — and how many searches that buys depends on how often each is checked. You can set a LOWER budget than your plan allows at any time. At these prices DataForSEO's $50 minimum deposit is the real cost rather than the usage, and it lasts a long time. Nothing is checked, and nothing is spent, until you choose searches to watch.
Choosing what to watch
Open Reports & Tracking → AI Overview and press "Add a search". You pick from the topics your projects already work on — your keyword topics and the ones topic discovery has found. The list is ordered strongest-evidence-first, and every recommendation shows the numbers behind it so you can disagree with it. WHERE WE GET THE EVIDENCE. We buy no keyword-volume database and estimate no volume — a number invented for a phrase and shown as a finding is exactly what we refuse to do. Where your project has Search Console connected we use something better: your own real queries from the last 28 days, with impressions, click-through and average position. Those are searches people actually ran to reach YOUR site, not a guess about what a phrase might get. THE STRONGEST CASE, and the one to act on first: real impressions, you sit on page one, and almost nobody clicks. That is what an AI Overview answering on your behalf looks like — people search it, they see you, and they never arrive. Watching it is the test of whether an overview is the reason. WHERE YOU ARE INVISIBLE there is no Search Console row at all, because you have never appeared for it. The evidence there is that a competitor holds page one: somebody in your market is investing in that search. That shows commercial interest, not demand, and it is marked as the weaker of the two rather than counted as equal. "NO EVIDENCE EITHER WAY" means exactly that — we know nothing about it, not that it is a bad choice. You may well know it matters; the badge is a statement about our data, not about your market. Marking a topic important on your project does not make it recommended. That is your opinion, and handing it back to you as a finding would be circular — it gets a mention and breaks ties, and it never changes the verdict. And if Search Console is not connected for that project, we say the evidence could not be reached rather than pretending none exists. Connect it and this list ranks itself on your own real demand. How many you can watch comes out of your budget, not an arbitrary number. Weekly checking makes each search cheap, so the list runs long; daily checking is seven times the checks, so the same money buys roughly a seventh of the list. The page shows both figures beside the toggle so you can see the trade before making it. There is one ceiling that is ours rather than your budget's — 500 searches per account. Every check is a separate request we make to DataForSEO on your behalf, and an unbounded list would monopolise the nightly run for everybody else. If you need more than that, talk to us and we will work out how to run it; raising your budget will not move that particular limit, and the page says so rather than sending you to change a number that would change nothing.
Reading the result
Each watched search shows one of three states, and they are three because they need different responses: • NOT CHECKED YET — no reading has come back. • NO AI OVERVIEW — Google shows no generated answer for this search at all. Writing a page to "win the overview" here is writing a page for something that does not exist. We re-check monthly rather than weekly, since this changes slowly. • NAMES YOU / DOES NOT NAME YOU — an overview appeared. This is the one worth acting on, and where it does not name you, the sources it used instead are listed underneath. That list is the answer to "why them and not us". Underneath each is a sentence with the counts: "showed an AI Overview in 3 of 4 checks, and named you in 1". Counts, not percentages — a percentage over four readings invents precision the sample cannot carry.
Weekly or daily — and what the choice costs
An AI Overview is generated, not retrieved, so two looks at the same search on the same day can genuinely differ. We proved that on our own account on 22 August 2026: one check returned a detailed answer that cited the customer twice; minutes later the same search returned nothing usable and cited nobody. Neither reading was wrong. That is simply how the surface behaves. Which is why a single reading is never reported as a fact — and why MORE readings are better, not worse. Seven a week give a firmer rate than one. So daily is offered, and what it costs you is list length rather than accuracy: the same budget buys about a seventh as many searches. Weekly keeps the list long; daily makes each search's rate firmer. Neither is the right answer for everyone, and the page states both numbers before you choose. A search where Google shows NO overview is re-checked monthly however often you check the rest. There is no point paying daily to be told repeatedly that nothing is there. Two kinds of reading are set aside rather than averaged in, and both are counted and reported so you can see where they went: checks that FAILED (an error is not an absence — a timeout does not mean Google showed no overview), and checks where Google REINTERPRETED your search into a different one. A different question is not another sample of the same question.
Which sites it keeps coming back to
Open any watched search and you get its source history: every site the overview has drawn on, how many readings it appeared in out of how many, and when it was last seen. Sites are labelled "keeps returning", "comes and goes" or "seen once", and your own domains are marked. That list is the actionable half. "You were not named this week" has no next step attached; "it has drawn on these four sites in every check for a month and you are not one of them" names the thing to go and be. TWO HONESTY RULES worth knowing about, because they change how you read it: Nothing is called "keeps returning" on fewer than three readings. On two, "in both" and "in one" are the only possible answers and neither is a pattern. And churn is normal. Published research puts the typical cited source's life at about four to five weeks, so most of what moves in that list week to week is the surface changing rather than anything about your site. The note under the list says so every time sources move, because reading ordinary volatility as your own decline is the one misreading this feature can cause.
Turning a finding into work
Where an overview appears and does not name you, the row offers "Work on this". It hands the search to the Evidence Engine as a topic on that project — the same route the competitor-gap rows use, with the same rules, so a topic added from either screen behaves identically. Most of the time it will come back "already a topic in this project's plan", because the picker offers your cluster topics in the first place. That is a real answer rather than a dead button: the engine is already building for it, and a second copy would not help. What it tells you is that the problem is not a missing topic — the evidence exists and is not being chosen — which is when the source list above becomes the thing to study. There is also "Check what's due now", for when you have just chosen your searches and would rather not wait for the nightly pass. It runs the same rules, so it cannot re-ask a search inside its re-check window; a button that could would let you spend your way to a worse answer.
What this does not tell you
- It does not measure Google AI Mode, or the ChatGPT app. Those are separate products with their own retrieval, and we do not claim them. - It is one search, from one location, once a week or once a day. It is a sample, not an index. - It cannot tell you how many people saw that overview. There is no impression data behind it. - Being cited is not promised by anything here or anywhere else. We measure; we do not guarantee.