Back to blogTechnical

Charleston Search Console: the query 'yes' ranks #6

Google logs AI Mode follow-ups as ordinary searches. I pulled two Charleston properties and found conversation turns ranking at position 6 with zero clicks.

7 min read·October 20, 2026·By Charlestowne Marketing

There is a query in one of my Search Console properties that is just the word "yes."

Eight impressions in the last 90 days. Zero clicks. Average position 6.3.

Nobody in Charleston typed "yes" into Google hoping to find a wildlife removal company. Somebody was talking to an AI, the AI offered to keep going, they answered it, and Google logged that answer as a search against my site. The report I use to judge whether the work is paying off now contains sentences that were never searches.

A study published last Wednesday says this is happening on most sites, and that it got roughly twenty times more common in sixteen months. So I went and checked my own.

What Google is actually doing

Google counts a follow-up question inside AI Mode as a brand new query. That is not a leak or a bug, it is documented behavior, and John Mueller confirmed in August that Search Console folds AI Overviews and AI Mode into the general performance report.

The discovery was not Google's announcement. An SEO named Anastasia Kourou noticed rows in her own report that did not look like searches ("yes", "yes go on", "yes, pricing") and asked Mueller about it directly. Barry Schwartz wrote it up on August 6, and Suganthan Mohanadasan published the full taxonomy a week later.

Here is the part that trips people up. Google does have a dedicated generative AI report in Search Console, added back on June 3. It shows your impressions and clicks inside AI features. It does not show you the queries. So the only place the conversational fragments surface is the ordinary performance report, unlabeled, mixed in with everything else.

Marie Haynes walked through that report the day it launched, and it is still the clearest explanation of what it does and does not give you:

The numbers behind it

On October 14, Jan-Willem Bobbink published a study of 155 Search Console properties, every query classified week by week from June 2025 through September 2026. Comparing the last eight weeks against the first eight:

  • Bare replies ("yes", "ja", "yes please") went from 5.57 to 116 per million impressions. That is 21 times as many, and they showed up on 94 of the 155 sites.
  • Mid-conversation pivots ("what about salary", "how about this") went from 0.12 to 4.56 per million. 39 times as many, with most of the growth landing after May 2026.
  • The median site's share of these queries went from 0.032% to 0.22% of all impressions. Seven times more.
  • In the week of September 21, 88 of 118 sites had them.

And the one that matters for anyone running a service business: these queries get clicked at 2.9%, against 35.6% for all queries pooled. Pivot follow-ups are the worst at 1.45%.

Bobbink is honest about the limits and I will repeat his hedge rather than strip it: this is inference, not measurement. Google does not label AI Mode queries. A bare "yes" could be a brand name or a person who genuinely types short. The trend matters more than any single row.

What it looks like on two Charleston properties

I pulled both of mine through the Search Console API this morning. The window is July 21 to October 19.

On the wildlife removal company I run here, three rows carry almost all of it:

  • "can you recommend cockroach control services in charleston sc?": 112 impressions, 0 clicks, position 17.3
  • "can you suggest a reliable provider for cockroach control in charleston sc?": 71 impressions, 0 clicks, position 18.1
  • "which companies in charleston sc offer mouse control?": 25 impressions, 0 clicks, position 18.1

Then the small, strange ones: "yes" at 8. "sure" at 2. "are they dangerous?" and "are they protected?" at one each, both sitting at position 6. "what about sc?" at position 9.

Read those last four out loud. "Are they protected?" is not a search. It is the second thing you say after somebody tells you there are bats in your attic.

In the previous quarter, April 20 through July 20, that property had nine impressions of bare replies and not a single one of the conversational questions. One quarter later it has 222 impressions I can confidently call conversation-shaped, which is 0.30% of the site's 73,697 impressions. The study's median site finished at 0.22%. A Charleston service business is already running ahead of it.

(My own agency site is smaller and the signal is thinner, but it has the single most interesting row of the bunch: "which charleston, sc marketing agency works with roofers and home services companies." Eight impressions. Average position 1.4. Zero clicks. I am winning that conversation and it has never once produced a visit.)

Not all of it is a human

Here is where I will disagree with most of the coverage I have read this week.

Those three cockroach rows are suspiciously tidy. Identical phrasing, repeated at volume, parked at the same position, never clicked. That is not 112 separate Charleston homeowners. That reads like an AI visibility tracker firing the same prompt on a schedule, which Bobbink's taxonomy treats as machine traffic and specifically does not count as conversational.

My agency property makes it obvious. Three of its queries end in the literal string "my location is charleston, south carolina, united states, usa." No human types that. Something is appending a location block to a prompt and sending it through.

So your performance report now holds three things that are not customers: real conversation turns, synthetic probes from monitoring tools, and pasted prompt fragments. All three add impressions. None of them add a phone call. If you have been puzzled by impressions climbing while the phone stays quiet, as I covered when Google answered a Charleston plumbing query with no web results at all, this is one more piece of the same shift.

The check, and it takes two minutes

  1. Open Search Console, Performance, set the range to the last three months.
  2. Click + New, then Query, then switch the dropdown to Custom (regex).
  3. Paste this: ^(yes|yeah|sure|ok|okay|more|go on|continue|show me|do it|thanks)$|^(what about|how about|and what about)\b|^(is it |are they |can you|could you|do you |tell me|give me)
  4. Set it to Matches regex and apply.

Whatever comes back is your conversation slice. Sort by impressions and read the actual strings, because the strings are the point. Repeated identical phrasing is a tracker. Short fragments and second-person questions are people.

What to do about what you find

Nothing dramatic, and I want to be clear about that, because the volumes are still tiny. This does not change what you build.

What it changes is how you read your own numbers. Average position across your whole site is now partly made of rows like "sure" at position 4.0, which is a flattering number attached to nothing. Site-wide CTR is now partly made of impressions that were never going to click. If you are judging an SEO program on those two blended averages, you are grading on a curve that keeps shifting underneath you. Segment the conversation slice out before you draw any conclusion, the same way I had to stop trusting a single weekly snapshot after five of nine Charleston results turned over in seven days.

The second thing is more useful. Those fragments are a transcript of what people ask after they have already started a conversation about your trade. "Are they protected?" is a page I should have. "Which companies in charleston sc offer mouse control?" is a service query I am being shown for at position 18, which is being shown and not chosen, which is the same gap I found when AI Mode started citing Maps listings instead of websites.

At this point somebody is thinking: 222 impressions, who cares. Fair. In isolation, nobody should. But it was 9 impressions a quarter ago on the same property, and the broad study says the curve has been going one direction since January. The useful move is not to act on today's volume. It is to know which part of your report is real before that slice gets big enough to fool you.

Run the regex this week. Three months of data, two minutes, and you will know whether anybody is talking about your trade to a machine in this market yet.

#seo-trends#search-console#ai-search#measurement#charleston#service-businesses

Want this kind of result for your business?

Free SEO audit. Two business days, no obligation, yours to keep.

Get a Free SEO Audit