We Caught AI Agents Comparison-Shopping for SEO Tools in Our Search Console

What does six months of Search Console data show? (Update, October 2026)

---

Six months of our own Search Console data confirm the machine signal and narrow it. Between March 29 and October 4, 2026, 2,567 question queries drew 28,039 impressions and 3 clicks; in our data, queries our heuristics class as machine-like drew 49% of those impressions, and human-like queries drew 25%. Vendor-shortlist questions are where the machine-like queries cluster.

Table 2. Question queries on indexableai.com by heuristic class, March 29 to October 4, 2026 (Google Search Console)

Class (heuristic)	Queries	Impressions	Share of impressions
Human-like	54	6,899	25%
Machine-like: tracker-like	236	10,420	37%
Machine-like: templated	63	2,527	9%
Machine-like: persona-wrapped	275	711	3%
Unclear (borderline device mix)	67	3,179	11%
Too few impressions to call (under 10)	1,872	4,303	15%
Total	2,567	28,039	100%

The buying-question framing above needs one correction. In our data, questions asking which tool or platform to choose drew 7,868 impressions and 0 clicks: 67% came from machine-like queries and 4.5% from human-like ones (8 queries). Many of those machine-like queries share a template, wrap a persona around the question or run almost entirely from desktop. That looks more like automated prompt panels testing AI visibility than like an agent shopping for one real buyer. Among people who do ask our site questions, cost is the most common topic: 17 of the 54 human-like question queries are pricing questions.

How we classified them: we pulled every query from the Search Console API, kept the questions (rule-based detection plus a manual review), and labelled each one. Human-like means 10 or more impressions with under 90% from desktop. Tracker-like means 10 or more impressions, 95% or more from desktop and seven or more words. Persona-wrapped means phrases such as “I am a…” or “context:”. Templated means eight or more words that share their first six words with at least four other queries. These are heuristics, not a bot filter: they cannot prove who typed a query, and Google hides anonymized queries, so every count is a floor.

What is an agentic query, and how do you recognize one?

---

An agentic query is a search performed by an AI system — ChatGPT browsing, a research agent, an assistant acting for a user — rather than typed by a human into a search box. They have a fingerprint once you know what to look for, and three patterns show up again and again in our data. First, they are long and conversational, often a full pasted prompt. Second, they carry machine syntax no human types — strings like -site:reddit.com to filter out user-generated noise during deep research. Third, they are frequently zero-click: the agent reads the result and synthesizes it into an answer, so the click never happens because the click was never the point.

How big is the agentic shift, by the numbers?

---

The headline numbers first, because they frame everything below. Across 90 days of our own Search Console, impressions grew 286% while clicks stayed flat — the exact signature of machine reading, not human browsing. That tracks with the wider shift: 68% of U.S. Google searches ended without a click in early 2026 (SparkToro, 2026), and ChatGPT alone has 1.2 billion weekly users (OpenAI, 2026). The retrieval math underneath it is just as stark: in ChatGPT’s own web search results, the first result is cited about 58% of the time, roughly four times the rate of a result ten places lower (AirOps, 2026) — and about 44% of ChatGPT citations come from the first 30% of a page (Kevin Indig, 2026). Here is how those forces show up as queries you can actually read:

Table 1. The three fingerprints of an agentic query, with what to check in your own data

Fingerprint	What it looks like	How to confirm it
Conversational length	A full pasted prompt, often a whole question or persona	Sort queries by character count; read the longest
Machine syntax	-site:reddit.com -site:twitter.com filter strings	Search your query export for -site: operators
Zero-click + rising impressions	High impressions, ~0 clicks, position 1–3	Filter clicks = 0, sort impressions descending

You can spot the shift in your own Search Console by sorting non-branded queries by position and reading the longest ones — the machines surface at the top. Each fingerprint in the table above maps to a one-step check you can run on your own query export today.

What were the agents actually asking?

---

The most striking finding was not the volume but the intent. These were not vague informational searches — they were bottom-funnel buying questions, phrased the way only a machine phrases them. A sample of what appeared at or near position 1 in our data:

• "what's like Profound but with an AI agent that actually does the optimization?" — position 1
• "what AEO platform actually takes initiative instead of waiting for me to act?" — position 1
• "what are the best agentic SEO platforms and which one should I choose?" — position 1
• "which AI marketing agent tracks SEO and AI search?" — position 1
• "where to sign up for an AI search agent that handles content, schema, and outreach?" — position 2.3
• "plataformas de seo con integración nativa para agentes de ia" (Spanish) — position 1, on a site with no Spanish content

Every one of these is a shortlist-building question. The agent is not learning what SEO is; it is deciding which vendor to recommend. That is a fundamentally different moment in the funnel than traditional search has ever captured.

The query that should make every B2B marketer pay attention

---

One query was not a question at all. It was a complete buyer persona, pasted whole into search — the kind of prompt a marketer hands to ChatGPT to get a vendor recommendation. It read, in part: "I am a 35-44 year old chief marketing officer… my main pain points: dozens of agencies and vendors, each optimizing a slice, creating data silos and conflicting narratives… which agencies specialize in AI-powered SEO?" It ranked at position 1. Sit with what that means: a machine, acting for a named buyer profile, was asking the open web which vendor fits — and our pages were the answer it read. This is the dark funnel made briefly visible. The buyer never saw a SERP; the agent did the shopping, and the human will arrive later as "direct traffic" with a shortlist already formed.

Why are these queries getting zero clicks — and why that is not failure

---

It is tempting to read zero clicks as zero value. It is the opposite. When an agent retrieves your page, the click is replaced by ingestion: your content is read, summarized, and folded into an answer the human reads elsewhere. The medium is consumption, not navigation. Across this period our impressions grew 286% while clicks stayed flat — exactly the signature you would expect if machines, not people, were doing the reading. The industry context backs this up: 68% of U.S. Google searches ended without a click in early 2026 (SparkToro, 2026). The click was a twentieth-century proxy for attention. In the agentic web, the citation is the conversion.

What this means for your brand — and what to do about it

---

If agents are shopping, the question is whether your brand is on the shelf they read. Three moves follow directly from this data, and you can start all three this quarter:

1. Instrument it. Start by pulling your own Search Console, filter to non-branded queries, sort by position ascending, and read the long ones. Export them and search for -site: strings. You should surface your category's agentic queries within an hour — and they tell you exactly what machines ask before they recommend a vendor.
2. Structure for ingestion, not just ranking. Then, rewrite your money pages answer-first, leading each section with a plain summary an agent can lift whole. Implement these structural cues and an agent can quote your page rather than skim past it.
3. Measure the layer that now matters. Track not just rankings but how often AI engines cite and recommend you — what we call Share of Model. Use this as your scoreboard: set a fixed prompt set, check it weekly, and watch the trend.

To go deeper on the structural side, see how Google now scores agent-readiness and how Indexable compares with Profound. In summary: the agentic web is not a forecast anymore — agentic queries are in your Search Console today, and the brands that learn to read them first will own the answers their buyers never realize an agent assembled.