Fundamentals · 8 min read · September 25, 2026 · Hafsa Tariq, gaash.ai
Which lanes AI answers name you for, and how current their facts are
AI visibility for a logistics provider is whether an assistant names it when a buyer asks about a lane, a cargo and a service level, a question that can come long before any RFQ, and whether what the answer says about its capacity and coverage is still true. A fixed panel of lane-and-cargo questions measures both, provided each answer's facts are recorded along with the name.
Before the RFQ, the list of names comes from somewhere
An RFQ goes to providers the buyer can already name. The names come from a colleague, a directory, a search engine or an assistant, and in the last case the list is written for one question: "which forwarders handle lithium batteries from Rotterdam to Jebel Ali by sea", or "who runs frozen freight from Montreal to New York twice a week". A provider left out of that answer is not on the list the buyer starts from.
Because the question names a lane, a cargo and a service level, so does the reading. A provider can be named for chilled freight to New York and absent from the same assistant's answer about frozen freight to Chicago, or named as a parcel carrier when it runs full truckloads. A logistics provider has no single AI visibility: it has a reading per service, lane and cargo, and one for its own name.
The probe beside our audit of 9,483 small business websites gives a starting point. One buyer question from each business's trade and metro went to Claude with web retrieval on, and 14.9% of 9,240 businesses were named. That is across all 40 trades, from one question on one day; we publish no logistics-only cut, and the figure describes no other assistant.
Named, recommended, first: what each reading means for a lane
Every answer can be read three ways, set out below, and our methodology builds its visibility score from the first alone: the share of the twelve questions that named the provider, with first place reported beside it. Named: the provider appears anywhere in the answer, under any spelling, in a list or an aside. Recommended: the answer puts it forward as a fit for the lane or cargo asked, with a reason beside it. First: it is the first provider the reader meets.
| Reading | In an answer about a lane | What to check next |
|---|---|---|
| Named | Your name in a list of providers for [lane] | What the answer says you do, and whether it is true |
| Recommended | "For weekly reefer loads to [city], [your name] fits because it runs [frequency] departures" | Which reason it gives, and whether a page of yours states that reason |
| First | Your name opens the list | The same question over three runs and four assistants before you believe it |
| Named for the wrong thing | "[your name], a parcel carrier" when you run truckloads | The page or listing that describes you that way |
Record the fourth row even though the score counts it as a mention. A name attached to the wrong service can bring RFQs you would decline, and can leave you off the list for the work you do run.
Why an answer's capacity and coverage can be years old
An assistant answers from what its model learned before its knowledge cutoff and, with web search on, from pages it fetches while it answers. Anthropic's documentation describes Claude's web search tool as giving Claude web content beyond its knowledge cutoff, with citations for the sources. Without a search, what an answer says about you rests on what was written about you when the training data was collected.
Logistics facts do not stand still. A warehouse opens or closes, a lane is added or dropped, a fleet grows, a certificate lapses or is renewed. An answer written from old training data, an old copy of a page or a directory entry nobody updated can state the old version as if it were current, and a buyer cannot tell from the answer alone.
The remedy is on your pages. Date every capacity figure and coverage statement ("[n] pallet positions at [site], counted [month year]"), send the pages of sites and lanes you no longer run to their nearest replacement with a 301 redirect, and correct the directory entries that still list them. When an answer states an old fact, the record shows the page it cited, which is where the correction starts; the article on wrong facts in answers covers the pages you cannot edit.
A panel built the way an RFQ is: service, lane, cargo
Write the panel the way the buyer writes the RFQ: by service, lane or region, and cargo, with the service level that decides the fit. That means twelve questions for a paying client, in the words your buyers use, left alone afterward so readings compare. The article on prompt monitoring covers the panel file and the discipline; the logistics part is choosing the questions.
| Question type | Bracketed example | What it reads |
|---|---|---|
| Service in a region | "which [service] providers cover [region] for [client type]" | Whether you are named for the service before any lane is given |
| Lane and mode | "who ships [cargo] from [origin] to [destination] by [mode], [frequency]" | The lane reading: named, recommended, first |
| Cargo and its conditions | "which carriers take [cargo] at [temperature range] on [lane]" | Whether the cargo you specialize in reaches the answer |
| Service level | "which [service] near [city] dispatches same day for orders placed by [time]" | Whether your cut-offs are what the answer repeats |
| Comparison | "is [your name] or [competitor] better for [lane]" | What the answer says separates you from the other name |
| Your own name | "does [your name] handle [cargo]" or "how big is [your name]'s warehouse in [city]" | What the answer states as fact about you, and whether it is current |
Weight the panel toward the first four types, which do not name you and are where a list of names starts. Choose the lanes and cargo you want more RFQs for, not every lane you run; a lane you add later goes in as a new line with its own start date.
What to record beyond the name
Each answer's record keeps the fields every panel keeps: named, recommended, first, who else, the verbatim answer, the date, the assistant and the run. A logistics provider adds four, because the facts in an answer can be wrong while the name is right.
| Field | What to write | What it catches |
|---|---|---|
| Service and lane named | The service and lane the answer attributes to you, in its words | Named for truckload when the question was about ocean freight |
| Certifications stated | Each certification the answer mentions, and whether you hold it today, from [date] to [date] | A lapsed certificate, or one you never held, stated to a buyer who will ask to see it |
| Capacity stated | Any figure for sites, pallet positions, fleet or throughput, beside your current dated figure | Capacity from a page, a listing or a year that is out of date |
| Page cited | The URL the answer cites, if any: your page, a directory, a competitor, a news item | Where each stale or wrong fact came from, and so where to fix it |
A "no" in the certification or capacity column is a task with an address: the cited page. Gathered over a month, those rows become a list of pages to fix, ordered by how often the answers repeat each one.
Reading a month of answers
| What the month shows | Where to look first |
|---|---|
| Named for your own name, never for a lane | Whether each lane you want has a page that states it; the specialist article covers writing one |
| Named, with a fleet or capacity figure years out of date | The page or listing the answer cited, then your own dated figures, as the operating figures article sets them out |
| Recommended on one assistant, missing on another | What that operator's crawlers can reach on your site; the comparison of AI search engines lists who fetches for each |
| A certification you do not hold, or no longer hold | The page or listing that states it, and your own certifications page; temperature-controlled work is covered in the cold-chain article |
| Every cited page is a directory | Whether your own pages state the fact at all, and whether a crawler can reach them |
Google's AI features give a second reading. Search Console's generative AI performance report counts impressions of your links in AI Overviews and AI Mode by page, country, date and device, with no query dimension. It shows which of your pages appear, not which lanes you were named for, so it sits beside the panel rather than replacing it.
What each page of a logistics site has to state, from services to certifications, is in the overview for logistics providers.
Common questions
Does AI visibility matter if our work comes through RFQs and tenders?
At the start of the process, yes. An RFQ goes to providers the buyer can already name, and an assistant's answer can be one of the places those names come from. A panel shows whether your name is in that answer for the lanes and cargo you want, and what the answer says about you when it is.
How many lanes should the panel cover?
The ones you want more RFQs for, not every lane you run. Ten to twelve questions split across service, lane, cargo and service level, with two or three that use your name, can be read month to month. Add a lane later as a new line with its own start date, so its history starts clean.
An answer gives our fleet size from three years ago. What should we do?
Read the page it cited, if it cited one. Put the current figure on your own page with the date it was counted, update or redirect the old page, and correct any directory entry that still carries the old number. Then ask the same question next month and record whether the figure changed. When no page is cited, the old figure may come from training data, and your dated page is the correction on offer.
Which assistant should a logistics provider measure first?
The one your buyers use, which new clients can tell you if you ask how they found you. If you cannot know, measure all four: each operator fetches with its own crawlers, so a provider can be named on one and missing on another. The 14.9% from our probe is a Claude figure and says nothing about ChatGPT, Gemini or Perplexity.
Sources
The probe figure is this site's own, from one question per business put to Claude once, across all 40 trades. The score is the one our methodology sets out, built from the named reading alone; what the web search tool and the Search Console report do is from their operators' own pages, linked where they are used.
Sources: our audit of 9,483 small-business sites, with its method · how the visibility score is measured · Anthropic, web search tool · Search Console Help, generative AI performance report
Read on