The AI-search demand your keyword tool reports as zero

Our site earns impressions for a family of AI-visibility queries. Search Console records every time Google actually showed our pages for these phrases since early May. Keyword Planner reports the same phrases like this:

The position column is what makes these numbers loud. Our pages sit at positions 59 to 82 for these queries, page six to page nine of results, and Google still showed them 2,700 times. An impression at position 82 only happens when someone pages that deep, so the queries' true search volumes are far larger than what a single page-eight site can witness. Keyword Planner reports them as zero.

These are not exotic phrasings. They are the natural words a marketer types after asking "does ChatGPT mention my brand?" Every keyword tool built on Keyword Planner data reports the opportunity as nonexistent.

The blind spot has a second, sneakier shape. Keyword Planner merges close variants into a single reported number, and it treats agent and agency as close variants. We pulled twelve months of monthly volumes for the pairs and compared them month by month:

"seo agent" reports 17,000 searches a month at a difficulty near zero. It looks like the opportunity of the year for anyone building AI agents. It is not: the volume belongs almost entirely to people shopping for an SEO agency. Anyone who targeted it off the tool numbers would write for buyers who are not there. The one genuinely distinct term in the set, "ai seo agent", is two orders of magnitude smaller.

None of this is a bug, and it is not unique to any one SEO tool. Keyword Planner exists to sell ads, and its data model makes four choices that are reasonable for ad buyers and misleading for anyone researching a fast-moving category:

• Twelve-month averaging. A term coined eight months ago carries months of zeros in its average, understating it by design.
• Volume bucketing. Reported volumes snap to a fixed ladder, so small-but-real demand rounds to zero or to "under 50".
• Close-variant merging. Distinct intents share one number, as with agent and agency, or singular and plural.
• Vocabulary lag. New phrases need to clear internal thresholds before they are reported at all. AI-search vocabulary is younger than those thresholds.

The result: the newer and faster-growing a search category is, the less keyword tools can see of it. AI search is the newest and fastest-growing category there is.

For established vocabulary, keyword tools remain useful. For emerging vocabulary, the honest instruments are different:

• Search Console impressions. First-party, unbucketed, and current. If Google shows your site for a phrase, the demand exists whatever any tool reports. Impressions on positions 50 to 90 are a map of demand you have not captured yet.
• Same-quarter comparisons. Compare a quarter against the same quarter a year earlier, never against the previous quarter, or seasonality reads as trend.
• Variant cross-checks. Before targeting any short keyword, compare its monthly series against its nearest variant. Identical numbers mean merged data, and the volume may not be yours to win.
• The live results page. Reported volume for an acronym is meaningless until you look at who actually ranks. AEO the marketing acronym and AEO the stock ticker share one volume number.