Search Console bot traffic is real, and on our own site it was big. Between July 14 and August 12, 2026, 96 machine-generated queries carried 22,637 impressions, 34% of rankzero.io's total, at positions our pages never held for a real search.
Those queries didn't just pad a number. They hid where our pages really ranked. Once we took them out, our main AI SEO agency page dropped from position 6.3 to 20.8, and our GEO service page from 9.0 to 50.7.
This post shows what the bot queries looked like, the filter our reporting now runs on every site we manage, why the obvious filter destroys real data, and how to check your own Search Console in about 20 minutes.
Key Takeaways
- Google's own team has said Search Console doesn't filter out all bot traffic. Rank trackers and AI visibility checkers type queries into Google, and every result they load counts as an impression.
- On rankzero.io, bots carried 34% of impressions in one 30-day window, concentrated on three pages at 92% to 98% each, and the average position on those pages was 14 to 65 places too good.
- A bot query has a shape: a question mark, 12 or more words, a number prefix, a search operator, or a stray brand token. It has impressions and zero clicks.
- Filter on shape plus zero clicks, never on length alone or on your own tracked prompts. Matching our own prompts would have deleted 419 real clicks across ten sites.
- The bots change. Three months later the family that made up our 34% had almost gone, and a numbered family had taken its place, so re-run the check every month.
What Bot Queries in Search Console Look Like
Bot queries in Search Console are searches no person typed. On our site they came in four families, and each one is easy to spot once you know the pattern.
The biggest family in August glued a real keyword to a misspelt brand name of another SEO tool, often with an emoji in the middle. Five spellings showed up. A few real examples from our query report:
ai seo for ecommerce 🐙 rankstr ioai seo for ecommerce try rankstr ioai seo agency for ecommerce rankstars iobest geo agency withrankstar io
We don't know who runs that bot. What we do know is how it behaves: it searches thousands of times for strings no buyer types, and it reads deep into the results.
The other three families are smaller but just as easy to spot:
- Numbered queries: A number, a colon, then a keyword:
11941: enterprise ranking analytics tool. It looks like a scraper working through a job queue. - Operator queries: Strings carrying
-site:reddit.comorinurl:, which almost nobody types while shopping for a service. - Prompt-shaped queries: Full questions like
how do i track my visibility in chatgpt?, the kind of prompt an AI visibility checker puts into Google to see what the AI Overview says.
That last family includes tools like ours. RankZero checks Google AI Overviews for the prompts our clients track, and any checker that does this leaves its prompts in someone's Search Console.
In the 30 days to November 2, none of the 18 prompt-shaped queries on our own site matched a prompt we track. They came from other people's tools.
Why Search Console Bot Traffic Ranks So Well, and What It Does to Your Report
A bot query ranks well because nobody else is competing for it. A made-up string like ai seo for ecommerce try rankstr io returns a near-empty results page, so whichever site matches a few of its words lands near the top.
That is why the damage goes beyond impressions. The bot inflates impressions, flatters your average position, and drags your click-through rate toward zero, all at once.
Here is what it did to five of our pages over those 30 days, raw figures first, then the figures with the bot queries removed:
- Our AI SEO agency page: 3,842 impressions at position 6.3. 98% were bots. Real: 86 impressions at position 20.8.
- Our GEO service page: 9,426 impressions at position 9.0. 92% were bots. Real: 796 impressions at position 50.7.
- Our ecommerce page: 7,130 impressions at position 9.9. 92% were bots. Real: 551 impressions at position 74.4.
- Our page for SEO specialists: 7,800 impressions at position 48.3. 5% were bots. Real: 7,418 at 49.8.
- Our AI search visibility glossary entry: 11,140 impressions at position 30.2. No bot queries at all, so raw and real agree.
The pattern matters more than any single number. The bots weren't spread evenly. They piled onto three pages and left the rest alone.
So any report that ranks pages by impressions, or flags "high impressions, low CTR" pages for a title rewrite, put the three most contaminated pages at the top.
That is what happened to us. In August we routed hire-intent work to our agency page because Search Console said it held position 6.3, our strongest result.
It didn't. It held 86 real impressions at 20.8, and the "quick win" list we had built in June and July was aimed at the wrong pages for the same reason.
If your Search Console numbers look better than your pipeline, this is the first thing we check. It is part of the measurement audit at the start of every engagement. Ask us for a measurement audit and we'll run it on your property.
Does Search Console Filter Bot Traffic?
No, not all of it. When someone asked about suspected bot traffic in the Performance report in 2021, Google's John Mueller replied, "Sometimes it can be from bots - we don't necessarily filter all of that out in Search Console."
The clearest evidence came in September 2025. Google stopped supporting the &num=100 parameter that rank trackers used to load 100 results in one request.
Across many sites, impressions fell sharply within days, and many site owners reported a better average position. Search Engine Journal and others read this as tracker impressions leaving the data. Google never confirmed the cause.
That change didn't end the problem. Our 34% was measured in summer 2026, almost a year later. Bots that load one results page at a time still count, and a bot query that ranks you on page one counts on every run.
How to Find Bot Impressions in Your Own Search Console
You can find most bot impressions with the Performance report's regex filter and one export. Steps 2 and 3 are the rule our reporting runs on every site we manage; Step 4 is the check we still do by eye.
Step 1: Pull queries, not pages
Open Performance, then Search results, set the date range to the last 28 days, and switch to the Queries tab. Page totals hide the problem, because a page with 98% bot impressions looks like a strong page.
Step 2: Flag the prompt shapes
Add a filter: Query, then Custom (regex), then Matches regex. Run these one at a time and note what comes back:
- Ends in a question mark:
\?$ - 12 words or more:
^(\S+\s+){11,}\S+$ - A number prefix:
^\d{3,6}:\s - A search operator:
(^|\s)-?(site|inurl|intitle):
Search Console uses Google's RE2 syntax, and a filter can be up to 4,096 characters, so you can join patterns with | once you trust them.
Step 3: Keep only what got impressions and no clicks
Sort the filtered list by Clicks, ascending. A flagged query only counts as a bot query if it has zero clicks and at least 20 impressions in the window.
Both conditions matter. A query with three impressions and no clicks is just a rare search. A query with clicks is a person, however odd it looks.
Step 4: Look for the families by name
Shape won't catch everything. Our brand-token family was short, under 12 words, with no question mark. Scan the top 200 queries by impressions for a token that repeats and makes no sense next to your keywords: a misspelt brand, an emoji, a domain fragment like io.
When you find one, write a pattern for every spelling of it. Ours needed five.
Step 5: Recompute without them
Export the query report to Google Sheets, drop the rows you flagged, and recompute impressions, CTR, and an average position weighted by impressions. Compare against the raw totals. That gap is your bot share.
Our GSC integration guide covers pulling the same query data through the API if you'd rather not export by hand.
Why "Long Queries With No Clicks" Is the Wrong Filter on Its Own
Every condition in Step 3 is there because a simpler filter failed when we tested it. Each one on its own either misses the bots or deletes real searches.
Length alone deletes buyers. Long queries are not always machines. On one consumer site we run, long queries convert at 14%. A rule that drops every query over 12 words would have thrown out some of that site's best traffic.
Matching your own tracked prompts misses most bots and deletes real clicks. It's the obvious fix if you run an AI visibility tool: remove any query that matches a prompt you track. We tested it in August 2026 across 10 sites over 30 days:
- On one site, it caught 2 of 50 prompt-shaped bot queries. The other 48 came from other people's tools.
- Across the 10 sites, it would have removed 419 real clicks. We deliberately track prompts that mirror what people search, so the prompt list and real demand overlap.
Shape plus impressions plus zero clicks held up. The same test flagged 66 queries and 2,350 impressions across the 10 sites and removed zero clicks. That last part is guaranteed by design: a filter that only takes rows with no clicks can't delete a click.
It fired where the problem was, at 30.5% of one site's query impressions, then 8.9%, 7.8%, and 5.5%, and not at all on five sites that had no problem.
On our own site it caught only 0.4%, because our brand-token queries were short and had no question mark. That is why Step 4 exists.
The Bots Change, So Your Filter Has To
A bot filter you write once goes stale. We re-ran the check on rankzero.io for the 30 days to November 2, 2026, three months after the August reading.
The brand-token family that made up most of our 34% had almost gone: 18 queries and 94 impressions. A numbered family had replaced it as the biggest one, with 414 queries and 2,263 impressions, and 2,171 of those landed on a single page, our enterprise rank tracker guide.
Across every family, bots carried 3,459 impressions, about 6% of the query report, and earned zero clicks. Six percent is a much smaller problem than 34%. But it's concentrated on one page again, which is exactly the shape that misleads a page-level report.
Read every share as a floor. Search Console hides rare queries from the query report for privacy, so the query table never adds up to the site total.
Ours covered 57,658 of 62,353 impressions this month. On some sites we manage it covers barely a quarter, and the bot rows hiding in the gap are never counted.
Put the check on a monthly calendar, next to your AI search tracking review. New families show up without warning.
What Clean Numbers Changed for Us
Filtering didn't make our report look better. It made it look worse, and that was the useful part.
The raw report said our commercial pages were our strongest assets. The clean report said they were our weakest, ranking on pages three to eight for the terms buyers type. That moved our plan from title rewrites, which can't earn clicks at position 50, to authority and links, which can.
It also changed how we read every client property. A page that suddenly shows position 4 on a term it never ranked for gets its query list checked before anyone celebrates. A CTR "opportunity" gets its bot share checked before anyone rewrites a title.
The same logic applies to AI traffic in GA4. If you want to see which sessions really come from ChatGPT and Perplexity, our AI traffic analytics separates them from the rest of your referrals.
FAQ
Does bot traffic in Search Console hurt my rankings?
No. Google has said a spike in impressions doesn't cause problems for search. The harm is to your decisions: inflated impressions and flattered positions send your team to fix the wrong pages.
Why do bot queries show high positions with no clicks?
Because the query is made up, so the results page is nearly empty, and your page ranks near the top by default. No person sees the result, so it never gets a click.
Are long, question-shaped queries in Search Console always bots?
No. Some are people typing AI-style prompts into Google, and long queries can convert well. Treat a long query as a bot only when it also has 20 or more impressions and zero clicks.
Did Google fix bot impressions in 2025?
Partly. When Google stopped supporting the &num=100 parameter in September 2025, impressions fell across many sites, which Search Engine Journal and others read as tracker traffic leaving the data. Our 34% was measured in 2026, so bots still reach Search Console.
Can I remove bot impressions from Search Console permanently?
No. You can't delete them from Google's data. You can filter them out of your own view with regex and recompute your numbers, which is what reporting should do every month.
How do I know whether my AI visibility tool is causing this?
Check whether the prompt-shaped queries match the prompts you track. On our site, none did. On the sites we manage, most came from other people's tools.
Run the Check Before the Next Report
Open your Search Console query report today and run the four regex patterns from Step 2. If the flagged rows carry more than a few percent of your impressions, recompute before anyone makes a decision on those pages.
If you'd rather have it done for you, a measurement audit is where every RankZero engagement starts: clean Search Console data, AI referral traffic split out in GA4, and a plan built on where your pages really rank. Book a strategy call and we'll show you what your Search Console bot traffic is hiding.