How to Spot, Handle and Report AI Agents Searching Google
.png)
Some search queries look less like something a customer would type and more like instructions accidentally left in the search box. When the same wording appears repeatedly, with only the brand name changed, it deserves a closer look.
These patterns can help you identify possible AI agents searching Google. They can't reliably tell you who sent them, or why.
The difference affects what content you create and how you report growth. A person asking an AI tool for help may have a real buying need. A monitoring tool running the same prompt every morning tells you much less about demand, however industrious it appears.
We checked our own Search Console and keyword data to see what we could separate. The patterns were easy enough to find. Explaining what they meant took more care.
Key takeaways:
- Repeated templates, pasted data and instruction-like wording flag possible automated queries, but none proves the source
- Long, chatty questions can come from real customers, so don't drop them for sounding like prompts
- When keyword tools disagree, check the history before you size demand
- Report suspect patterns as their own group, because filters can't give a clean count of human demand
Can You Spot AI Agents Searching Google?
You can spot queries that deserve a closer look, but you can't confirm that an AI agent sent them. Search Console doesn't label a query as human or software, and keyword tools don't either.
We use "AI-shaped queries" for searches worded like a prompt or an automated template. It's a working label, not a confirmed source. An agent can also search on a person's behalf, so the people and software groups won't always separate cleanly.
What to look for in Search Console
Three patterns in our data pointed to possible automated searches.
- Pasted rows and comma-separated data. One query is a spreadsheet header row. Another pairs a keyword with six numbers, as if pasted from an export.
- Repeated templates and instructions. We found many versions of "evaluate the [category] company [brand] on [facet]," with the structure fixed and the subject swapped. Others read like orders, such as "provide a definitive answer."
- Odd, repeated endings. Several queries end with ". my location is usa." A person could type that, but the same ending across many searches points to a template.
A detailed, chatty query may state exactly what a buyer wants to know, so length alone is a poor reason to drop it.
Google counts a follow-up question in AI Mode as a new query, with its impressions and clicks credited to that question.
Our post on conversational search queries covers how to read these searches.
How to group them in one view
Add a Queries filter in the Performance report, choose Custom (regex), and pull the suspect patterns into one view.
Start with this filter, built from our own data, and swap in the templates you see:
(,|evaluate the|my location|provide a definitive answer)
The bar means "or." Choose Matches regex to inspect these queries, or Doesn't match regex to see the rest. Google's regex filter guide explains the options.
A comma is a useful search aid, not a confession, and the queries your filter misses aren't proven human either.
What Did We Find in Our Data?
Our data shows a few clear AI-shaped patterns, but they cover only a small share of the impressions.
We looked at tenpointlabs.com web search from April 1 to October 9, 2026. Impressions rose more than 40-fold between April and September, while clicks rose less than two-fold. Average position slipped from page two to page three.
That gap started our review, but it doesn't prove an AI explanation, since we published new pages in the window.
The baseline also needs care. Google documented an impressions logging error from May 13, 2025 to April 27, 2026. It affected impressions, click-through rate and average position, but not clicks, and most of our April baseline sits inside that window.
The clearest patterns covered a small share of impressions:
- Queries with commas made up 1.6% of impressions in the window and earned no clicks
- "Evaluate the," "my location" and "provide a definitive answer" added less than 0.4% between them
Those figures describe what our filters caught, not the full share of automated activity. Some matches may be human, and automated queries with ordinary wording slip through.
Search Console also leaves anonymized queries out of the table. A query filter drops them from the chart totals too, so a filtered view won't add up to the unfiltered report.
What Did the Keyword Tools Show?
The keyword tools showed similar odd patterns, and the second tool we checked didn't back them up. On October 12, 2026, Ahrefs listed dozens of near-identical tracker-style keywords, each at 90 to 1,200 searches a month with no cost per click. DataForSEO returned no data for the 11 we tested.
Seven of the Ahrefs keywords end in "my location is usa." and were first seen on October 4 or 5, 2025. The same ending appears in more than a dozen of our own Search Console queries.
One phrase that pairs a brand name with a product category shows a 12-month average of 1,100 searches a month in Ahrefs, against 42 in the latest month, and DataForSEO again returned no data. A gap that wide may mean a short burst, so check the monthly history first.
Ahrefs combines several sources to estimate volume, including Google Keyword Planner, Google Trends and third-party data, so we can't rebuild any single estimate.
Our conclusion is narrow: the wording is unusual, a second tool doesn't back these keywords up, and their demand is unconfirmed. Calling them fake, or blaming a named agent, would go past the evidence.
How Should You Handle AI-Shaped Queries?
Use the wording to decide what to check, then judge what each query tells you. This table sorts the common cases:
A recurring prompt that names your brand may mean someone is monitoring you. It doesn't tell you who, whether they're a buyer, or whether your brand appeared in the final AI answer. So use software queries as evidence of visibility, never of demand.
Our guide to measuring AI share of voice shows how to use prompt tracking that way.
How Do You Plan Around AI-Shaped Queries?
Keep suspect keywords out of your demand numbers until they pass a few checks. Before one shapes a brief or a forecast, run these five:
- Read the query in context. Does it describe a real customer problem, or look like a repeated instruction?
- Compare a second tool. Match the location and other settings. If one tool shows 1,000 searches a month and another shows nothing, record the disagreement.
- Check the cost per click. A missing cost per click on a commercial keyword is a flag, especially next to other odd signs.
- Look at dates and monthly trends. A group of similar first-seen dates may show how a tool added keywords to its database.
- Check your own Search Console. Look for the query and close variations. A query missing from your data still doesn't show a lack of demand.
Add an "Other evidence" column to your planning sheet, and note whether a second tool, your own data or customer research backs each doubtful keyword.
Put keywords with one volume estimate and nothing else on a watch list. Check your own tools too, because a rank tracker or AI visibility tool you run may be the sender.
Once a keyword passes, our B2B keyword research guide shows how to plan content.
How Should You Report AI-Shaped Queries?
Report them as a separate, labeled group, never as growth in demand. Build a view that shows the patterns without presenting them as proven AI activity. A good report does four things:
- Shows the filter and its limits. Call the group "AI-shaped queries" and say which patterns it covers
- Shows impressions next to clicks. When visibility rises much faster than visits, say so before anyone calls it growth
- Keeps the overall totals. Show the filtered view beside the original report, with a note that a query filter drops anonymized queries
- Notes the known gaps. If your chart spans May 13, 2025 to April 27, 2026, flag Google's impressions error and how it affects the baseline
Google's Generative AI performance reports began rolling out to a subset of sites in June 2026. They list impressions, pages, countries, devices and dates but no queries, so use them for page and date views, not for sorting queries.
If you report on AI visibility, our guide to citations versus mentions explains which counts belong in a report.
How we do this at Tenpoint Labs
We compare the keywords we plan around in Ahrefs and DataForSEO, then check the client's Search Console where it has data. A keyword that only one source shows, or that has volume but no cost per click, goes on a watch list before it reaches a plan.
The data behind it
The main view is Query intelligence in Tenpoint Analytics. It shows which queries are close to page one and where each one sits in the buying journey. Our strategists read it next to the Page performance view, which shows clicks and rankings.
A templated query with impressions and no clicks gets reviewed before it steers a plan.
What you see in your seat
You see the same two views we plan from, and we connect it for you, so you can ask Claude or ChatGPT which pages earn impressions but no clicks. That one question shows how much of your traffic story rests on visibility alone. See what's in a seat on Tenpoint Analytics.
FAQs
What are AI-shaped queries?
AI-shaped queries are searches worded like a prompt or an automated template. Some come from people chatting with AI, such as follow-ups in AI Mode, and others may come from software, such as scheduled prompts from AI visibility tools. Search Console labels none of them.
Can I filter AI-shaped queries out of Search Console?
You can't filter them out cleanly. Search Console doesn't label queries as human or software, so you filter by shape: commas, templates such as "evaluate the," and location endings. Our comma filter caught 1.6% of impressions, and some matches were ordinary questions, so use the filter to label groups and report them apart.
Do AI-shaped queries hurt my rankings?
We haven't seen any sign that they hurt rankings. They can distort reports, though: extra impressions with no clicks pull click-through rate down, and our average position slipped as impressions grew. We can't say how much of that slide came from software.
Should I target a keyword that only one tool shows?
You shouldn't target a keyword on one tool's volume alone. Check its cost per click, a second tool and your own Search Console first. Write for it only if one of them agrees.
Start with one Search Console filter this week. Review the matching queries, note their share of impressions, and list the patterns that need a closer look. You won't find every AI agent searching Google, but you can stop a suspect volume estimate from becoming next month's content plan.
