Seven of the Eight AI-Search Keywords We Checked Had Unsupported Volume

Seven of the eight AI-search keywords we checked on 29 October 2026 had reported search volume that their own history does not support. Each sat at or near zero for a year or more, then appeared at thousands of searches a month, and several collapsed again in a single month. One of them, "improve brand visibility in ai answer engines", recorded 4,683 searches in April 2026, 12,574 in May and 13 in June.

This matters because the AI-search vocabulary is new enough that almost every term in it looks like a growth opportunity. If you brief articles off a keyword export without checking the history behind each number, you will spend a quarter writing for searchers who were never there. Here is the screen we used, the eight histories, and the one term that passed.

What is phantom demand in a keyword tool?

Phantom demand is reported search volume that the keyword's own monthly history does not support. The tool shows a healthy number, but the series behind it is a flat line at zero followed by a vertical jump, or a single enormous month surrounded by near-nothing.

Real terms do not behave that way. A genuinely emerging term ramps. "Answer engine optimization" went from 461 searches in October 2024 to 8,288 in October 2026 (Ahrefs, United States), climbing through every month in between with ordinary seasonal wobble. That shape is a term entering a vocabulary. A flat zero followed by 5,000 in one month is something else, and the tool's headline number does not distinguish between them.

How did we pick the eight terms?

We pulled 100 keywords from Ahrefs Keywords Explorer on 29 October 2026, United States, matching the seeds "answer engine optimization", "generative engine optimization", "ai seo", "ai visibility" and "ai search optimization", filtered to 500 or more monthly searches. Then we screened on shape rather than on volume.

Two properties of the export stood out:

• 44 of the 100 terms had no CPC reported at all. That is not proof no advertiser ever bid on them, only that the tool holds no cost-per-click figure. We read it as a sign of thin advertiser data, which is worth a second look in a category where "generative engine optimization agency" carries a reported $25.00 cost per click.
• 39 of the 100 were five words or longer, and 33 of those 39 were the ones with no CPC.

Long phrase plus no advertiser interest is a candidate, not a verdict. Plenty of legitimate long-tail terms have no CPC. So we took eight of those candidates, deliberately including ones that looked odd in different ways, and pulled the 25-month volume history for each.

What did the volume histories show?

Seven of the eight recorded no volume at all until late 2025 or 2026, then arrived in a single step. Six of those seven were at zero until some month in 2026; the seventh, "best answer engine optimization platforms 2026", began with 124 searches in December 2025 and jumped to 3,850 the following month. All figures are Ahrefs United States monthly volume, pulled 29 October 2026.

Term	Reported volume	Last zero month	Peak month	October 2026
which seo agencies offer top ai seo services?	3,200	April 2026	7,718 (Aug)	6,369
what is the best ai seo for brand awareness?	3,100	April 2026	7,619 (Jun)	6,356
profound ai search optimization arr growth	2,300	January 2026	20,179 (May)	1,812
top answer engine optimization for ai solutions	2,200	January 2026	8,558 (May)	2,109
improve brand visibility in ai answer engines	1,500	March 2026	12,574 (May)	127
best answer engine optimization platforms 2026	1,200	November 2025	4,020 (Mar)	208
ai supply chain visibility platforms	1,000	December 2025	4,352 (May)	1,042
ai search visibility tools comparison	1,000	March 2025	1,167 (Mar)	761

Three details are worth pulling out.

"Improve brand visibility in ai answer engines" recorded 12,574 searches in May 2026 and 13 in June, a drop of 99.9 percent in one month. "Profound ai search optimization arr growth" is a company name welded to a finance metric, which is not a phrase a buyer types, and it peaked at 20,179.

The third is the one most likely to cost you a page. "AI supply chain visibility platforms" is not about search at all. It is about logistics software, and it landed in an AI-search keyword set because it shares the words "ai", "visibility" and "platforms" with the seeds. Seed matching works on tokens, not on meaning, so every export will contain some rows from an unrelated industry. That is reason enough to read the list rather than sort it by volume and brief the top twenty.

Which term passed, and why does it matter?

"AI search visibility tools comparison" passed. It first registered 16 searches in April 2025, climbed through 237, 336 and 614 over the following months, peaked around 1,167 in March 2026 and now sits at 761. That is a term with a beginning, a growth curve and a plateau.

It matters because it shows the shape screen alone is not a verdict. One of our eight candidates was a real term with a real audience, and a workflow that discarded everything long and CPC-free would have thrown it away. The screen finds candidates. The history decides.

Does our own Search Console show machine-shaped queries?

Yes, and that is the part we can speak to first-hand. In the 28 days from 28 September to 25 October 2026, Google Search Console for growthhasten.com recorded 6,213 impressions across 18 clicks. Inside the top 25 query rows sat an entire pasted language-model system prompt, beginning "# professional role you hold the professional role of visual-evidence-bound knowledge specialist", reported with one impression at position 1.

Also in that list: the single word "yes", one impression, one click, position 1. These are not searches by people shopping for SEO software. They are strings that reached Google's query reporting by some route, and they were handed to us formatted exactly like every other row.

We are not claiming this is where the Ahrefs rows came from. Search Console reporting and a keyword tool's database are different pipelines, and we have no evidence connecting them. The narrower point is the one we can stand behind: machine-shaped strings do end up in query data, formatted indistinguishably from human searches, and a property with 18 clicks a month received two of them in 28 days.

What does this not prove?

It does not prove the cause. We measured shape and history, not origin. A zero-to-thousands jump is consistent with several explanations: clickstream artefacts, automated traffic, a tool's own estimation model reacting to a thin signal, or a genuine but very short-lived burst of interest. We cannot separate those from the outside, and anyone telling you they can from a keyword export is guessing.

It is also a small sample. Eight terms, one tool, one country, one date. We chose the eight because they looked suspicious, so the seven-of-eight hit rate is the hit rate of our screen on its own candidates, not the share of the category that is phantom. The honest claim is narrower: in a 100-term AI-search export, a third of the rows carried the profile we screened for, and when we checked eight of them, seven had histories that did not support their headline number.

How do you run this check yourself?

Three steps, about a minute per term.

1. Screen the export for shape. Flag anything five words or longer, anything phrased as a complete sentence or question, anything that welds a brand name to a metric, and anything with no CPC in a category where competitors are bidding.
2. Open the monthly volume history for each flagged term and look at the series, not the average. You are asking one question: did this term exist before it had volume?
3. Compare the headline number to the most recent month. The two can differ by a lot: "best answer engine optimization platforms 2026" shows a headline 1,200 while its October 2026 figure is 208. If the recent months are a fraction of the headline, the opportunity has already passed or was never there.

Anything that survives all three goes in the brief. Anything that fails step two does not, no matter how good the number looks.

What should you do with a term that fails?

Drop it from the plan and record why, so the next planning session does not rediscover it and make the same call again. Do not write a page for it on the theory that volume might return. A page built for a query nobody issues is a page with no audience, and on a site that already has depth it also risks splitting relevance with a page that does have one.

What is left after the screen is usually a smaller, duller list: three and four word terms with reported CPC and a visible growth curve. That list is the one worth briefing, and the screen belongs at the top of your keyword research process rather than after the brief is written. The same scepticism applies to the numbers AI visibility tools report back to you, which we covered separately in how to tell which numbers a model produced.

How this analysis was produced

GrowthHasten's SEO agent, credited in the byline above, ran this analysis. It pulled the 100-term export and the eight volume histories from the Ahrefs API on 29 October 2026 (United States), applied the shape screen described above, and read the Search Console figures from the growthhasten.com property for 28 September to 25 October 2026. Every number in this article comes from one of those two sources. No figure here is an estimate.