When Google shows an AI Overview, the top-ranking result loses 58 percent of the clicks it would otherwise take (Ahrefs, May 2026). So your visibility dashboard can keep climbing while those clicks, and the revenue behind them, fall. Fix your attribution before you buy a share-of-answer tracker, because that gap is where the money leaks. Across the market as a whole, 68 percent of US Google searches in the first four months of 2026 ended without a click to any website (SparkToro, June 2026), which means a large part of your market builds a shortlist before anyone clicks.
Share-of-answer trackers measure being cited, not being recommended
Being cited is not the same as being recommended, and the trackers sell you the first while you need the second. Profound, Otterly.AI and Peec AI each report the percentage of prompts in which an engine mentions you. As a rough direction of travel, that has some use. But a brand is often quoted as a source without being the brand the engine then recommends, so the score the playbook optimises for is not the outcome you are paying for.
A tool that samples chatbot answers never touches your Search Console or your revenue, so it cannot see where your buyers went. Your problem is not a missing share-of-answer score. It is that your existing reporting cannot show where the demand actually formed.
Your dashboard can rise while your traffic falls
Appearance rate counts how often an AI Overview shows up, not how often it costs you a click, and the two are constantly confused. Ahrefs, in research published in May 2026 comparing Search Console click-through across 300,000 keywords before and after AI Overviews arrived, found the top-ranking page loses 58 percent of its clicks when one is present (Ahrefs, May 2026). Whole-market clickstream data points the same way: SparkToro's June 2026 study of US Google behaviour, built on Similarweb panel data, found 68 percent of searches between January and April 2026 ended without a click to any website (SparkToro, June 2026). The two methods differ, but both point one way: a growing share of your category's questions get answered on the page, and the visit your reporting was built around is no longer guaranteed.
The click that does happen no longer means what it used to. A buyer can read an AI Overview naming three suppliers, form a shortlist, then later type your competitor's name straight into the browser. Your analytics records that as nothing, as direct, or as branded search you claim credit for after the fact. Google makes it harder to catch: AI Overview clicks and impressions are counted inside the aggregate Web search type in the Search Console Performance report, with no dedicated filter to isolate them (Search Console Help: Performance report). Take those totals at face value and you look more visible while your real traffic slides underneath.
Treat share-of-answer as a quarterly signal, not a weekly KPI
Any share-of-answer score is directional, because the answers underneath it do not sit still. Large language model responses are non-deterministic: OpenAI's documentation states that lowering the temperature setting makes output more deterministic, and that at higher settings the same prompt returns different completions (OpenAI). The same question can return different recommendations from one session or phrasing to the next. That property tells you what the number is fit for. It answers one question: have we entered or left the recommended set in a category over a quarter. It cannot tune content week to week.
To separate signal from noise, define one metric and hold to it: your prompt inclusion rate, the share of your audit prompts in which the engine names you. Run each prompt several times in a single audit and treat the run-to-run variation as your noise floor. One prompt flipping between named and unnamed is noise. A shift across the whole set, sustained from one quarter to the next, is signal. As an illustration only, using a sample of fifteen audit prompts rather than the whole category: moving from named in three of those prompts to six, an inclusion rate rising from 20 to 40 percent, is the kind of change worth acting on. Anyone selling this as a precise, week-by-week metric is overselling it.
That raises a fair objection: if the score only moves quarterly, what does a client pay to change week to week? The answer is the inputs, not the index. Entity resolution, page clarity and publishing quotable benchmarks are weekly work with weekly output. You ship the causes on a weekly cadence and read the effect a quarter later. What you do not do is refresh a number every Monday and call it progress. Anyone promising that is selling you the noise floor as if it were performance.
You get named by publishing something worth quoting
Engines name the businesses whose content they can read, trust and quote, not the ones with the longest keyword list. Take a services page that states everything and asserts nothing, and replace it with one that says plainly what the firm does and who for. Resolve the firm's name consistently across the web so engines treat it as one business rather than several. Then publish a single benchmark drawn from your own delivery data. When the AI Overview answering the category question has a specific, attributable number to quote and no competitor offers one, it quotes yours and names you as the source.
Your paid budget should follow the decision, not the click
Paid systems optimise to the clicks and conversions they can see, which is exactly the moment AI answers now hide. Smart Bidding and Performance Max chase measurable conversions, so when a buyer's shortlist forms inside an answer with no click, that moment is invisible to the algorithm and budget drifts towards demand it can still capture. Amazon built its advertising business on the opposite principle, selling placement on the pages where people research and choose, and reported 19.8 billion dollars of advertising revenue in the second quarter of 2026 alone (Amazon Q2 2026 results, July 2026). It monetises the decision, not the click-out. The lesson for paid media is to widen the frame from cost-per-click to presence at the decision, and to do it only once your attribution is honest, because a broken picture keeps rewarding the wrong spend.
Run a prompt-set audit and reconcile it against Search Console this quarter
Run a prompt-set audit before you commit any budget. Write down the ten to twenty questions a real buyer asks when they enter your category. Put each to ChatGPT, Gemini, Perplexity and Google's AI Overviews. Record whether you are named, whether the description is correct, and whether an AI Overview is present at all. Repeat it monthly from the same location so movements mean something.
Then reconcile it against Search Console and isolate the cause. Falling clicks on their own will not tell you why, because rank drift, seasonality and page layout also move click-through rate. Split your queries into the set where an AI Overview is present and the set where it is not. If click-through rate falls on the AI Overview set while the other set holds and your average position is stable, AI answers are taking the visit, not rank or season.
A fair challenge: that split proves correlation, not that the AI Overview caused the lost visit, and the shortlist-then-direct path is invisible by design. Close the loop with corroborating signals. If branded search volume for your name rises while clicks on the AI Overview set fall, buyers are meeting you in the answer and coming back by name. If direct-traffic share climbs fastest in your AI Overview-heavy categories, the same pattern is surfacing in your own analytics. No single line proves causation. Branded search lift and direct-traffic share moving together with the click gap is as close as this channel lets you get, and it is enough to act on.
Two questions belong on the board's dashboard alongside sessions. Are we named when the engines answer for our category. Are we described correctly when they do. If either answer is no or unknown, that is the gap between what you spend and what you earn.
