The Local Blind Spot

The pattern is inverted

Grouped by the shape of the query. The more local the phrasing, the less the AI metric sees. The one shape it reports generously is the one with no location in it at all.

Median monthly Google volume per shape, with the AI figure beside it.

And it overstates the other end

The metric is not simply conservative. On 7 queries it reports a number higher than Google's, and every one of them is a generic head term with no location. Median AI volume for that shape is 2,851 against a Google median of 3,900.

This is the shape of a metric built from question corpora rather than from search behaviour. Generic questions are the ones that get written down, quoted and answered on the open web. “Plumber near me” is typed, acted on, and never discussed.

How we measured

The rules below were written before the data was collected, and the query list was generated from templates rather than hand-picked, so that no one has to take our word for which examples we chose.

The comparison

The same query list was sent to two DataForSEO endpoints on the same day: ai_optimization/ai_keyword_data/keywords_search_volume for the AI figure and keywords_data/google_ads/search_volume for Google's own advertising data. United States, English.

The query list is generated, not chosen

12 categories crossed with 4 templates: “X near me”, “best X”, “X in CITY” and “best X in CITY”, across 8 US cities. 120 queries in total. The templates are the shapes our own product generates, not shapes picked to produce a gap.

Null is not zero

Google returned no data at all for 24 of the 120 queries. Those rows are excluded from every figure on this page rather than counted as zero, because counting them as zero would flatter the AI endpoint. All totals use the 96 rows where both endpoints returned a number.

What a zero means

A zero from the AI endpoint is a real response, not a failure: the endpoint returns 0 for queries it has no signal for. That is the point of the finding. It is reporting zero demand for queries where Google measures hundreds of thousands of searches.

Cost

$0.11 in total: $0.02 for the AI figures, $0.09 for the Google figures. The whole study is reproducible for the price of a coffee, which is part of why we think nobody should take an unaudited volume number on faith.

Reproducing it

Collector: backend/scripts/study-local-blind-spot.ts. The published dataset carries all 120 rows including the ones we dropped.

What this does not tell you

• Google volume is not AI volume. Nobody knows how often people ask an assistant for a plumber, and this study does not claim to. It says the numbers being sold as AI volume do not track the demand we can measure, and that they fail hardest exactly where local businesses compete.
• One vendor, one endpoint. We measured the source we actually use. We did not test every AI volume product on the market, and this is not a ranking of vendors.
• Google's own numbers are estimates too. Keyword Planner volumes are rounded and bucketed. They are a far better instrument for local demand than what we compared them against, not a perfect one.
• One country, one language, one day. US, English, a single collection pass. No trend claim is possible from it.
• A zero is not proof that nobody asks. It is proof that the metric has no signal there. Treating a missing measurement as an absence is the mistake this whole page is about, and it would be a poor way to read it.