Answer engine optimization and search engine optimization are not two different content jobs. I checked mine. Google’s generative AI report says 218 pages on this site were shown inside AI features 5,350 times over the last 90 days, and nine of the ten most cited are pages that already earn organic clicks. The work that gets you quoted is the work that already ranks. What changes is what you are allowed to measure, and that is where the budget is about to get wasted.
An AI answer once sent me a paying client, so I am not the sceptic here. But I spent this morning joining two Search Console tables on the URL column, and the result is duller than anything I have been pitched.
The AI list was not a different list. It was the same list, about twice as long, and sorted differently.
What is actually new here, and it is not the data
Google’s help page says that as of 31 August 2026 these insights were rolled out to all websites worldwide. Two weeks ago. In my own console today it is headed Performance on Search generative AI features and still carries a Beta label.
Here is the part almost nobody says out loud. The data is not new. Google’s documentation changelog entry for 16 June 2025 is headed AI Mode is now counting towards totals in Search Console, and the help page for the new report says it includes data from the Web search type. So these impressions have been inside the total you report to clients for fifteen months.
The report did not start measuring anything. It drew a line around a slice you already had. Which means the AI number sits inside your Web number, and adding them together counts the same impressions twice.
The mistake: buying AEO as a second content programme
The pitch writes itself. AI search is a new channel, so it needs new content, new markup and a new retainer line.
Google’s own documentation settles the markup half: its guidance on AI features says there is no special schema.org structured data that you need to add. There is no AI row in the search appearance dimension, no AI filter in the main performance report, and no field in the bulk export. No third party tool is rebuilding this from an API, because there is no API value to rebuild from.
That leaves the content half, the expensive half. It is answerable with data, because you already hold both lists. Pull them.
AI search is a new channel, so it needs its own content programme and its own retainer line.
Buys a second content track before anyone has checked whether the two lists differ.
Export both tables, join them on the URL, and count how many pages sit in both.
On my site the answer was 79 pages, 71 per cent of every click earner. There was no second track to buy.
What my own 218 cited pages say
Ninety days, 12 June to 11 September 2026, on imisofts.com.
The generative AI report: 5,350 impressions across 219 rows, 218 unique pages once you normalise the duplicate URL forms. The ordinary performance report over exactly the same window: 366 clicks, 110,000 impressions, average position 14.6, and 111 pages that earned at least one click.
Join them on the URL and this falls out. Nine of my ten most cited pages also earn organic clicks. Sixteen of the top twenty, twenty three of the top thirty. Seventy nine pages sit in both lists: 71 per cent of everything that earned a click, carrying 71 per cent of all my AI feature impressions.
The top of that list is not a set of exotic AI ready assets. It is a best-of comparison, a scraping tools roundup and four cold email law pages, all written for people.
I set the bar low on purpose: earning a click means one click in ninety days. The low bar is the finding. The cited pages are not a separate population, they are drawn from the one that already works.
The page cited 668 times that has never been clicked
Now the part that should stop you celebrating.
My most cited page drew 668 impressions inside AI features in ninety days and earned zero clicks. Not a poor click through rate. Zero. And 139 of my 218 cited pages, 64 per cent, earned nothing at all.
So the honest version of my finding is not the tidy one. The AI list is not a different list. It is the same list, about twice as long, sorted differently. Your winners are in it. Most of it is not a winner.
Being shown inside an answer is not demand. Report 5,350 to a client as reach and you are reporting a number that has never once converted on your own site.
Four things the report will not tell you
Four limits, before anyone builds a dashboard on this.
It is impressions only. Four dimensions: pages, countries, dates and devices. No query dimension and no click column, so you will never learn which question you answered.
Position inside it means nothing. Google states that an AI Overview occupies a single position in search results and that all links in it are assigned that same position. An AI Overviews rank tracker is therefore selling you one number that every source in the answer shares.
It undercounts you. Google says two results from your site in one AI response count as a single impression in the chart total.
You cannot read a trend off it yet. My ninety day chart climbs steeply through August, which looks like a wonderful story. Two things say do not tell it. The report only reached all sites on 31 August, and Google’s own data anomalies page records a logging error that cut impressions on this exact report from 13 to 17 August, restored on 21 August. My figures postdate that restore, so the totals are sound. The curve is not something I would put in a deck.
Four moves this week
No tool required. One morning and a spreadsheet.
Export both tables on the same day
Open the generative AI report, export the pages tab, then export the pages tab of the ordinary performance report over an identical window. Same dates or the comparison is worthless.
Join on the URL, never by eye
Strip trailing slashes and the console’s inline button text before you match, or your join will silently miss rows. Mine reported four of my top ten pages as uncited until I normalised the slugs. The real answer was nine.
Score the overlap before you approve a budget
Count how many of your click earning pages appear in the cited list, and what share of AI impressions those pages carry. Two high numbers mean your existing format already works on both surfaces, and AI search optimization is a volume and measurement job, not a rewrite.
Stop adding the AI number to your traffic number
It is a slice of the Web search type, so adding it double counts. Report it on its own line, label it impressions, and put a how did you hear about us field on your booking form to catch what it cannot show you.
The bottom line
If your cited pages and your click earning pages are broadly the same set, you do not have an AEO content problem. You have a volume problem on the format that already works, and a measurement problem on a number you were handed two weeks ago.
That is a cheaper diagnosis than the one being sold, and it is the one I reached about my own site after building this for 200+ businesses. What I would spend money on instead is what closed the loop for me: a how did you hear about us field on the booking form. Search Console will never tell you an AI answer sent someone. One form field will.
When I run AI search optimization for a client, this join is the first deliverable, on their data, before anybody writes a word. Half the time it kills the retainer they came in asking for. That is the correct outcome.