My AI Agent Fleet Post Is Ranking for Truck Fleets. A Term-Collision Postmortem.

On 2026-08-02 I pulled the full Search Console query list for campilii.com through the API — every row, including the zero-click ones nobody looks at. The property is small enough that I can read all of it: 18 queries in the 28-day window ending 2026-07-30. Twelve of those 18 contain the word fleet. Roughly half of them are not about software at all. They are about trucks.

This is the postmortem of a search-term collision: I named a real thing with a term another industry already owned, and Google is now quietly telling me — in the position column — exactly which half of that term I actually serve.

The raw numbers

Pulled from the Search Console API on 2026-08-02 for the property https://campilii.com/, date range 2026-07-02 to 2026-07-30, dataState: all, rowLimit: 500 — so this is the complete query table, not a top-N view. Google defines an impression as a user having "seen (or potentially seen) a link to your site in Search, Discover, or News," and position as the ranking of that link, documented in the Search Console help.

Every "fleet" query in the window, sorted by average position:

Every one of those rows resolved to a single URL: my definition post on what an AI agent fleet is. In the same window that page shows 23 impressions and an average position of 36.7 — which is the weighted blend of a page doing reasonably on one meaning of a word and badly on another.

One honest caveat about the arithmetic before anyone checks it: the query table for that window totals 24 impressions, while the page table totals 79. That gap is not an error. Google omits rare queries from the table entirely to protect user privacy — "anonymized queries," which stay in the totals but never appear as rows — and Search Console's own troubleshooting documentation for data discrepancies says so plainly. On a property this small, most individual searches are rare searches. I can see the shape of demand here; I cannot see all of it, and I am not going to pretend otherwise.

The collision is real, and it is not my fault or Google's

Two industries genuinely use this vocabulary, right now, in public.

The vehicle side: Motive's page on AI agents for fleet management describes "AI agents" as autonomous systems that monitor vehicles, drivers, and logistics workflows. That is a large, well-funded, commercially mature category — telematics, compliance, driver safety, trucks.

The software side: the NHIMG glossary entry for "agent fleet" defines it as "a managed population of AI agents or semi-autonomous systems operating inside an organisation," framed as an identity and permissions problem. No vehicles anywhere.

Same two words. Completely different buyer, budget, and body of knowledge. When somebody types "ai fleet" into Google with no other context, there is no correct single answer — the engine has to guess, and it hedges by showing pages from both senses and watching what people do.

I knew this when I wrote the original post; it already carries a section on the two meanings. What I did not know until I read the query table is how the collision would score. That turned out to be the interesting part.

Average position is a free intent classifier

Look at the table again with the labels covered. The software-sense queries land at 12, 16, and 32. The unambiguous vehicle-sense queries land at 33, 56, 66, 83.5, and 86. There is essentially no overlap between the two bands.

That is not Google being confused about my page. That is Google being extremely clear: for the meaning you actually cover, you are a mid-first-page-to-page-three candidate; for the meaning you do not cover, you are page six filler that got shown because the string matched. Nobody had to tell me which posts serve which intent. The ranking distribution did it, at no cost, from data I already owned.

This generalizes past my odd little vocabulary problem. If a page ranks 60+ for a family of queries and 15 for another family, the position spread is a measurement of topical fit, and it is more honest than any keyword tool's relevance score, because it is the actual engine's actual judgment about the actual page. I found this only because I pulled the entire query list rather than the top rows — the diagnostic signal was living entirely in the zero-click, one-impression tail that a dashboard summary rounds away. That is the same lesson as the six-brand audit where Search Console was broken on five properties: the data you never look at is where the finding is.

What I am deliberately not doing

The tempting move is obvious. Vehicle fleet management is a bigger market with more search volume than "how do I run several Claude Code agents on a schedule." I am already being shown for those queries. Writing four posts on AI for DOT compliance and real-time telematics would probably lift something.

I am not going to, for three reasons, and the third one is the real one.

I would be fabricating expertise. I have never operated a truck fleet, never touched an ELD, and have no first-hand data about driver-safety scoring. Everything I could write on it would be a synthesis of other people's pages. My entire position is that I write from systems I actually run — the moment I publish a confident guide to something I have only read about, the value of every other post drops, because a reader can no longer assume the numbers are measured.

It would be scaled content, structurally. Publishing into an adjacent vertical purely because the impressions are there is the pattern Google's guidance treats as unhelpful content produced primarily for search engines. I have watched this play out on another property in my own fleet: a domain with 1,150 submitted URLs is being held back by roughly 507 thin pages, and the fix there is deletion, not more publishing. Adding wrong-vertical pages to a 34-post site would be doing that on purpose.

Those impressions are not my customers. Every one of these posts exists so that a specific reader — someone deciding whether to hire an AI systems builder — finds evidence before the conversation starts. An impression from a logistics manager searching for telematics software is not a lead; it is a number that makes a chart look better. Optimizing for it would trade the thing I want for a metric that resembles it.

So the answer to "you are ranking 86th for fleet compliance management" is: correct, and that is the appropriate position for a page that has nothing to say about it.

What I am doing instead

Three things, all narrow.

Sharpen the sense I do serve. The queries in the 12-to-32 band — "what is a fleet agent," "ai agent fleet," "agent fleet" — are the ones worth an inch of effort, and the gap on that page is not the definition, it is the operational detail underneath it: how the agents are scheduled, what the shared doctrine file actually contains, what happens when one silently fails. That detail is what I have and the glossary pages do not. It also happens to be what my other posts on running a fleet across five brands already document.

Disambiguate on the page, for the reader, not the crawler. Someone who lands from "ai fleet" expecting telematics should be able to tell within one sentence that they are in the wrong place and leave. That is a courtesy, and it is also the cleanest signal I can send about what the page is.

Keep pulling the whole list. Monthly, all rows, both windows. The collision was invisible in every summary view and obvious in the raw table.

The micro-lesson

Before you build a content position on a term, search it and read the first two pages as if you were the person typing it. If another industry already owns the phrase, you are not competing for a keyword — you are competing for a meaning, and you will be measured on both meanings whether you wanted the second one or not. And once the data arrives, resist the specific temptation it creates: impressions from the wrong audience feel like traction and are the cheapest thing in the world to accumulate. The position spread across an ambiguous term is one of the most useful free diagnostics you own. Read it, then act on the half you can honestly serve, and let the other half sit at 86.

FAQ

What is a fleet agent?

The term has two live meanings in 2026. In software, a fleet agent is one member of a managed population of AI agents — scoped, permissioned, and governed as a group; the NHIMG glossary defines an agent fleet as "a managed population of AI agents or semi-autonomous systems operating inside an organisation." In transport and logistics, the same words describe AI systems that monitor vehicles, drivers, and compliance for a physical fleet. Nothing in the phrase itself tells you which one is meant, which is why search engines serve both.

Is an AI agent fleet the same thing as AI fleet management?

No. AI fleet management is a vehicle and logistics category — telematics, driver safety, route and compliance automation for trucks and vans. An AI agent fleet is a set of individually scoped software agents sharing a written operating doctrine, a scheduler, and verification gates. The overlap is a word, not a concept, and the vendors, budgets, and skills involved have nothing in common.

Why does my page rank for queries from a completely different industry?

Because a string matched and the engine had nothing better to serve for that specific phrasing. This shows up as impressions at deep average positions — in my case 56 to 86 — which is Google's way of saying the page is a last-resort candidate for that intent. It is generally harmless. It becomes a problem only if you mistake those impressions for demand and start writing toward them.

Should I write content for a search term my industry shares with another one?

Write for the meaning you can serve from first-hand experience, and let the other meaning go. Chasing an adjacent industry's version of your keyword means publishing about work you have not done, which is both a credibility risk and structurally the kind of search-first content Google's helpful-content guidance is designed to suppress. On a small site, four wrong-vertical posts can be a meaningful share of the whole domain.

How do I see the zero-click queries Search Console hides in the summary view?

Pull the Search Analytics API directly with dimensions: ["query"], a high rowLimit, and dataState: "all", rather than reading the dashboard's top rows. That returns every query row Google will release. It will still not be all of them — rare searches are omitted as anonymized queries to protect privacy, so your query totals will come in lower than your page totals, which Google documents in its data-discrepancy troubleshooting page.

— Italo Campilii. Every number above was pulled from the campilii.com Search Console property on 2026-08-02 for the window 2026-07-02 to 2026-07-30.