Published September 22, 2026Measured September 17–22, 2026
Google's AI Overview cites no one
on local searches
For six consecutive days we read the AI Overview block for 22 queries, once a day, through the same API and geolocated to the same towns. Every one of the 68 reads produced an Overview. On the 23 reads whose query named a town, the Overview showed zero source links — all 23. On informational queries in the same window, it showed four to nine. This page publishes the whole table, our own zeros included.
01 — The record
Sixty-eight reads, one method, one window
Between September 17 and 22, 2026, we read Google's AI Overview once a day for 22 queries: some are market queries for local businesses in Broward County, Florida, and three are our own. Every read was pulled the same way — the Overview block of the result page through a SERP API, geolocated to the business's own city. Every one of the 68 reads generated an Overview, so nothing here is an absent answer counted as a zero.
Grouped by the shape of the query, the result splits cleanly.
| Kind of query | Queries | Reads | Sources shown in the Overview |
|---|---|---|---|
| Queries naming a town or city “pilates davie”, “gym weston fl”, “andropause treatment in davie” |
13 | 23 | 0 sources, every single time |
| Informational and comparative queries in English “hormone replacement therapy”, “f45 vs orangetheory”, “generative engine optimization agency” |
6 | 31 | 4 to 9 sources in 25 of 31 reads |
| Brand queries, and queries in Spanish “agencia de marketing con IA”, “August marketing agency beaugust” |
3 | 14 | 0 sources, every single time |
Source: August's own measurement panel. One read per query per day, same API, same geolocation, September 17–22, 2026. “Sources” counts the distinct domains the Overview block showed as a citation. No client metrics are published here: the queries are market queries, not brand queries, and no client site is named.
02 — What it means
An answer with no sources is not a ranking you can lose
The usual way to read a zero is “we were not chosen.” On these 23 reads nobody was chosen. The Overview answered “pilates davie” and “gym weston fl” and “andropause treatment in davie” without pointing at a single page, ours or anyone else's. There was no list to be on.
That changes what the work is. On a query that names a town, the block behaves less like a citation surface and more like a summary Google writes for itself — and the attention that leaves it lands on what sits underneath, which on a local query is the map pack. So the asset that decides the outcome is the Google Business Profile: categories, hours, photos, reviews, and a site whose structured data does not contradict any of it. That is unglamorous work and it is the work that pays on those queries.
The other half of the table says the opposite thing, and it is just as useful. Take the town out of the query — “hormone replacement therapy”, “f45 vs orangetheory”, “what is f45 training” — and sources come back, four to nine of them, in 25 of 31 reads. Those are the questions where a page can still be cited, and they are asked by the same person earlier in the same decision.
Which gives a content rule that is easy to apply: write the citable pages against the questions people ask without naming a place, and let the profile carry the ones where they do name it.
03 — What this does not say
The limits of six days and two towns
Twenty-three reads on thirteen queries, in one metro area, over six days, with one method. That is enough to call the pattern consistent and not enough to call it a rule about Google. If your market is somewhere else, the honest move is to run your own queries for a week and write down what the block did.
The local group is also not the only one that came back empty: the three brand and Spanish-language queries returned zero sources on all 14 reads. So “names a town” is not the only shape that produces a sourceless Overview, and this page is not claiming it is. It is claiming that on the local shape the behaviour did not vary once.
There is a separate observation in the same record that deliberately does not appear in the table above. On September 18, 2026, run by hand with web search enabled from the same two towns, a conversational engine answered local queries of this kind by naming specific studios. Different source, different method, three runs per query instead of one. It is worth knowing that the two surfaces behave differently — and its numbers are not put next to the ones above, because readings from different sources do not compare. Mixing them would produce a chart that looks like evidence and is not.
How answer engines choose which sources to cite · Our own domain, measured daily with the zeros in
04 — What to do with it
Three moves, in this order
First, measure. Pick the eight or ten queries a customer would actually type to find a business like yours, half of them with your town in them and half without. Read the Overview once a day for a week and write down two things: did it generate, and did it cite anyone. That is a baseline, it costs nothing, and without it no one can ever prove to you that they moved something.
Second, put the profile first on anything that names your town. Claimed, categorised correctly, hours that match the door, photos of the inside, reviews answered. On the 23 reads above, nothing else was available to win.
Third, keep the writing for the questions with no place name in them. What a treatment involves, how one method differs from another, what to expect the first time — those are the reads where sources still appeared, and a page written to answer one of them plainly is the only kind that can be pulled into an answer.
And a test that runs itself, for any agency selling you visibility in AI answers: ask for their own measurements. Not a client case — theirs. Three of the queries in the table above are ours, and they returned zero sources fourteen times out of fourteen. That is published here on purpose.
05 — Questions
What people ask about this
Does Google's AI Overview never cite sources on local searches?
That is not what this record says, and the difference matters. It says that in 23 out of 23 reads, across 13 different queries naming a town, over six consecutive days in September 2026, measured the same way from the same place, the AI Overview was generated and showed zero source links. That is a strong, consistent local observation — not a law about Google. Two towns in Broward County, Florida, one metro area, one measurement window. Run it in your own city before you assume it holds there.
If the AI Overview cites no one, is local SEO pointless now?
The opposite, but the target moves. When the AI Overview names no sources, the click does not go to a cited page — it goes to whatever sits under the block, which on a local query is the map pack. So the surface that still converts is the one that was always there: the Google Business Profile, its categories, hours, photos and reviews, and a site whose structured data agrees with it. What stops paying on those queries is the work aimed purely at being cited inside the answer, because on these 23 reads there was nothing to be cited into.
Why publish measurements where your own site scores zero?
Because three of the 22 queries in this window are ours, and they returned zero sources on all 14 reads. Leaving them out would have made the whole table a selection instead of a measurement. An agency that only shows you the readings that went well is showing you a shortlist.
How were these numbers measured?
One read per query per day of the AI Overview block in Google's result page, pulled through a SERP API and geolocated to the business's own city, between September 17 and 22, 2026. Sixty-eight reads, 22 queries, one method. Every one of the 68 generated an Overview, so no missing answer is being counted as a zero. Anything measured another way — ChatGPT with web search, for instance — is kept separate on this page and its numbers are not placed next to these, because readings from different sources do not compare.
What should a local business do about this?
Three things, in this order. Measure your own queries for a week and write down whether the Overview cites anyone — it costs nothing and it is the only baseline that will ever be yours. Make the Google Business Profile the priority on any query that names your town, because that is where the attention lands when the answer cites no one. And keep the citable work for the questions people ask without a place name — what a treatment is, how one method compares to another — because those are the reads where sources still showed up.
06 — Keep reading
The rest of the log
Pieces from the same measurement series, held to the same rule: every figure with its date and its method.