So lately, we have had a few conversations with clients around the strange happenings in Google Search Console. Impressions are up, clicks are flat or down, and somewhere in the query list there is something like:
“I am a 45-54 year old vice president for enrollment management at a mid-sized private university and I need to understand how peer institutions measure brand equity.”
Who in the world is searching for such a specific query?
There is an answer. Or, more accurately, a few answers.
Query fan-out, AI Mode, synthetic prompts, and automated searches are all changing what shows up in the data we use to understand organic search. The Search Console performance report still counts impressions, but in 2026 those impressions do not all represent the same thing.
Here is how we have been sorting them.
In 2026, the job is increasingly to separate what was measured from what the metric actually proves.
What is query fan-out, and why is it in your Search Console?
Query fan-out is what Google’s AI Mode does with a question. Someone asks one thing, the model breaks it into several sub-searches, runs each one, reads the results, and writes a single answer. Every one of those sub-searches can register as an impression for whatever pages it read. Google has also confirmed that a follow-up question inside an AI Mode conversation counts as a brand new query, which is why you will occasionally see fragments like “yes, go on” in your report. That is literally someone’s reply to the AI.
The people behind those queries are real. They are asking about you. They are just never going to see a blue link, so they are never going to click one. An impression with no click from query fan-out is not a failure. It means your page was read and, in some cases, quoted. That is the same mechanism behind the university in our AI Overviews case study, just seen from the other side.
Visibility is not demand.
Three kinds of traffic
Typically, when we pull the query export for a .edu right now, we see three broad types of activity mixed together.
The first is the humans.
Real prospective students, parents, and staff typing familiar searches: a program name, “tuition,” your school plus “apply,” or a question about financial aid.
These queries tend to be shorter, they have click-through behavior you would recognize, and they are what most higher education dashboards were originally designed around.
They are still there. They are just sharing the report with a lot more noise.
The second is AI-assisted search activity.
These are longer questions, conversational searches, follow-ups, and sometimes fragments that look more like part of a conversation than a traditional Google query.
These can represent real users interacting with AI Mode, and they can still generate clicks. But they also create a different kind of visibility than traditional search, so I would not interpret them the same way I interpret a conventional program or admissions query.
Most higher education dashboards still have one line labeled 'organic search,' and that line no longer means one thing
The third is synthetic and automated activity.
AI visibility platforms run constructed prompts to see which brands, vendors, universities, or institutions get mentioned. Agent tools can submit long instruction-style queries. Scrapers, automated systems, and other tools can generate searches that look surprisingly human.
The persona example at the top of this post is a good example of the kind of query we would investigate.
We cannot say with absolute certainty that every query like that is a bot. But when you see dozens of highly structured persona prompts built around presidents, marketing VPs, enrollment leaders, or other job titles, they are much more consistent with synthetic monitoring activity than normal search behavior.
And this matters because a query being visible in Search Console does not mean there is meaningful audience demand behind it.
A university appearing for 200 synthetic 'What school should I choose?' prompts is interesting. It is not the same thing as 200 prospective students searching for that university.
Visibility is not demand
This is where AI tracking gets especially messy.
A growing number of platforms measure whether a university or company appears when they run hundreds or thousands of synthetic prompts.
Those tests can be useful for research. What they cannot tell you reliably is whether real people are asking those same questions at meaningful scale.
There is no clean equivalent of traditional keyword search volume for AI prompts. Responses can vary from one session to another. Conversation history and personalization can change the answer. And a citation or brand mention does not automatically prove that the cited page caused the answer or influenced the user’s decision.
Rand Fishkin recently made this same argument when discussing the limitations of the AI visibility tracking industry: these tools can produce useful directional signals, but synthetic prompt coverage, citation counts, and “brand visibility” percentages should not be confused with actual audience demand or business impact.
That distinction matters a lot in higher education. A university appearing for 200 synthetic “What school should I choose?” prompts is interesting. It is not the same thing as 200 prospective students searching for that university.
How to tell them apart in about ten minutes
You do not need a new platform for this. In the Search Console performance report, filter to the last three months, export queries, and sort by word count. Anything under six words is mostly human. Anything over ten is mostly not, and anything that opens with “I am a” or “as a” or contains a full sentence about the searcher’s job title is a bot probe. In between is where AI Mode conversations live: grammatical questions, “what about” follow-ups, replies with no subject.
A regex filter on the query field catches most of it. It will not be perfect. Some long queries are real people who type in full sentences, and some short ones are bots. But it gets you from “traffic is down and we do not know why” to “human search is roughly flat, AI Mode visibility is up, and here is what is noise.”
What Google’s new Generative AI report does (and what it does not tell you)
Google’s Generative AI performance report launched in June 2026 and separates AI Overviews and AI Mode impressions from classic search, which is useful. It also shows no queries, no clicks, and no position for those impressions. It tells you that your links appeared inside an AI answer and nothing about what the AI did with them. Treat it as a visibility count, not a traffic report, and do not expect it to answer the bot question, because it will not.
What to do with each bucket
For humans, nothing changes. Keep measuring away.
For query fan-out and AI Mode traffic, look at which pages are being read. If the AI is pulling from a 2021 PDF instead of your current program page, you may want to remove it. Make sure you have clear headings, one obviously definitive page per subject, an answer in the first paragraph, structured data. It is the same list from our AISEO is just SEO post, whic is a great reference.
For the bots, exclude them before you build anything on the data. That is it. Do not chase them, do not write for them, and do not let them inflate the impressions number you show leadership, because someone will eventually ask why impressions went up 40 percent and applications did not.
The reporting problem
We think the real issue is not simply that Search Console got noisy.
It is that most higher education dashboards still have one line labeled “organic search,” and that line no longer means one thing.
When we build an SEO reporting dashboard for a client now, it is increasingly useful to separate:
- traditional human search behavior
- AI-assisted visibility
- synthetic or automated activity that should not drive strategy
AI visibility platforms can live alongside that reporting too, but I would treat them as diagnostic tools rather than audience measurement tools.
A synthetic prompt test can tell you something interesting:
“Google or ChatGPT frequently associates our institution with nursing.”
What it cannot tell you by itself is:
“Prospective nursing students are increasingly interested in our institution.”
Those are two totally different claims.
AI visibility can also fluctuate even when nothing meaningful changed about your website. Different prompts, different conversation histories, personalization, and model variation can all change the result. That makes a single “AI visibility score” very different from something like a search ranking or an organic session.
Leadership does not need to understand all the mechanics of query fan-out.
They need to understand what the metrics actually represent.
The humans are still coming.
The institution may increasingly appear inside AI-assisted search experiences.
And some of the scary-looking impression growth may simply be machines talking to machines.
Some of the scary-looking impression growth may simply be machines talking to machines.
Impression growth is no longer synonymous with audience growth
For years, an increase in Search Console impressions generally meant one thing: Google was showing your pages to more searchers.
That assumption is becoming less reliable.
An impression may still represent a traditional search. It might also come from an AI-assisted search experience, a conversational follow-up, or automated activity that has very little relationship to your actual audience.
That does not make Search Console useless.
It means we have to get better at interpreting what it is actually measuring.
The same applies to AI visibility platforms. They can be useful.
The mistake is treating every impression, citation, mention, or synthetic prompt as evidence that your audience grew.
In 2026, the job is increasingly to separate what was measured from what the metric actually proves.
If you want to try the sort yourself, the export takes about ten minutes.
Take a look at what share of your top fifty queries appear to represent normal human search behavior, conversational AI activity, or something synthetic.
You may find that the story behind your impression growth looks very different once those things are separated.
And if you would rather have us look at it with you, contact us. We have a Data Visualization and Reporting Strategy team who would love to discuss reporting options and optimization strategies.
Frequently Asked Questions About AI Overviews and AI Visibility
Query fan-out is the process where Google’s AI Mode splits one user question into multiple sub-searches, retrieves results for each, and synthesizes a single answer. Each sub-search can create a Search Console impression for the pages it reads, usually without a click.
Three common reasons in 2026: your page was read by AI Mode during query fan-out and cited or summarized without a link being clicked; a bot or AI visibility tool ran a synthetic query that surfaced your page; or your page ranks for a query where the human intent is answered on the results page itself.
Partly. The Generative AI performance report, launched June 2026, shows impressions from AI Overviews and AI Mode but no queries, clicks, or positions. AI Mode conversation fragments also appear as ordinary queries in the main performance report.
Export your queries, sort by word count, and flag anything over ten words that reads as a persona description, an instruction, or a pasted string. A regex filter on first-person openers and job-title language catches most synthetic prompts. Exclude those before reporting trends.
