There is a way to lose a job that never reaches your desk.
A facilities director has a system failing and a budget conversation starting. Rather than opening a search engine, she describes the problem to an assistant and asks who handles this kind of work near her. Back comes a paragraph and three company names. She calls one of them. Nothing was bid and nothing was declined, and nobody at your firm will ever hear that the project happened at all.
The reflex is to read that as a ranking problem, or as evidence that the market has gone strange. It is neither. It is a supply problem. The system assembling that answer had nothing about your firm solid enough to say out loud.
That is diagnosable, and the causes are a short list.
What has to be true before your firm gets named
Nobody outside the companies that build these systems knows exactly how an answer gets assembled. What you can observe is the output, and it is consistent enough to work backwards from.
Three things have to be true before your name can appear: the material has to be reachable, it has to be readable as plain claims, and those claims have to agree with what an answer engine finds elsewhere. The third is the one almost nobody builds for.
That third condition accounts for most absences with a traceable cause. These systems reliably prefer a vague answer to a wrong name, and that preference is visible in what they return. Offered a firm they can describe from several agreeing sources and a firm they can only partly resolve, they name the first and drop the second. The better contractor loses to the better-documented one.
So the useful question is not how to rank in ChatGPT. It is narrower and more answerable: what does this system currently have on file about us, and is any of it sturdy enough to repeat?
Five reasons a strong firm comes back absent
Nothing you have published is quotable. Most technical firm websites are built to present rather than to answer. A capabilities page is a run of adjectives and a services grid is a taxonomy; neither one contains a sentence a machine can lift. If nobody at your firm has ever written a plain answer to something your buyers genuinely want to know — what a replacement inside an occupied building involves, how a shutdown window gets planned, which scope you carry with your own crews and which you don’t — then there is nothing to extract, however good the work is.
Your basic facts are not machine-readable. Structured data is labeled code inside the page that states what a person would otherwise infer from the layout: legal name, trade categories, territory, hours, who runs the firm. Without it, every one of those facts is a guess, and nothing else on the web can agree with a guess.
The firm does not resolve to one clear entity. Firms fracture themselves without noticing. A DBA on the trucks, a different legal name on the license, an old suite number on a directory, a second phone number on the maps listing, a professional profile describing a company the website does not quite match. Each mismatch gives the system one more reason to treat everything about the firm as unconfirmed.
Nothing outside your own site vouches for you. Agreement needs a second voice: an association roster, a permit record, a manufacturer’s dealer page, a trade publication, a client’s own announcement of a project, a complete maps profile carrying recent reviews. A firm can hold an excellent reputation with almost no public paper trail — common enough in industrial and commercial trades — and that is exactly the profile that comes back absent. A reputation held entirely inside people’s heads is unreadable from outside.
Sometimes it is only the door. One small file at the root of your website tells automated visitors which of them you are asking to stay out. A past developer, a security plugin, or a hosting default may have set it to turn away the ones that feed answer products. Worth knowing: the crawlers gathering training data are named on separate lines from those retrieving pages for live answers, so you can decline one and welcome the other. Check this first. It does not touch your conventional search rankings or what a human visitor sees, but leaving it wrong keeps an answer engine from ever reading the site the other four fixes live on.
What your public record has to look like in 2026
None of this is an AI project. It is the ordinary work of maintaining a public record that matches the business, which is what being found means — the first stage of a growth system, sitting ahead of being trusted, chosen, or measured.
The conclusion most owners get sold at this point is volume — more pages, more posts, more activity. That is the wrong read. Publishing additional material that says nothing specific gives a system nothing new to work with. What works is narrower and considerably duller than a content campaign.
One page per real question, answered in the opening lines. Not a blogging habit. A small set of pages, each taking one thing a buyer wants to know before they contact anybody, and answering it directly at the top rather than after four paragraphs of preamble. A handful of those does more than a long run of general posts, because each is a complete answer somebody can lift instead of a point buried inside a page that is really selling something.
Say identical things everywhere. One legal name, one address format, one phone number, one description of what the firm does and where it works — repeated without variation across your site, your maps listing, your professional profiles, and any directory that carries you. This is unglamorous, takes an afternoon, and returns more than anything else on this list, because it is exactly the kind of agreement these systems reward.
Put a named person behind it. A stranger — and a system summarizing for one — can do more with a named author who has a verifiable background than with anonymous corporate text. For an owner-led firm that is an advantage rather than a burden: you already have the license, the years, and the project history. They are simply not written down anywhere a stranger can reach them.
Be concrete about territory and scope. Sweeping claims are unverifiable by design. Naming the corridors you work in, the facility types you handle, and the systems you are actually credentialed on gives a retrieval system something it can check — and gives the buyer the same thing.
Give one person the job of keeping the record true. Not a marketing function. A half-hour each quarter confirming nothing has drifted: a changed number, a relocated yard, a service line added two years ago and still mentioned nowhere.
Taken out of order, that sequence produces very little — which is the larger argument in why technically strong firms lose ground they never see.
Three checks worth running this week
Ask an assistant to describe your firm by name. Ignore whether it is flattering. Look at whether the trade, the coverage area, the size, and the specialties are right. Wrong details are more useful than no details, because they point at which stale source is being leaned on — and a bad source is easier to correct than an empty record.
Ask for the sources, not just the answer. Run the open question a prospect would run, then ask where those names came from. That list is a free and unusually honest competitive read: the pages doing the work in your market, named for you.
Look at your own site the way a machine does. Fetch your homepage as plain text, or view it with scripts turned off. If the substance of the page only arrives once a browser runs code, some retrieval systems see a nearly blank document. Counterintuitively, the risk is highest on sites built as single-page applications, because heavy front-end frameworks are exactly what delay the text.
The part worth holding on to
Absence from an AI answer is not a verdict on your work, and it is rarely mysterious. It means the record is thin, inconsistent, unconfirmed, or unreachable — and each of those has a specific fix that belongs to you rather than to a platform.
The firms named a year from now will not be the ones with the most content. They will be the ones whose public record is complete and consistent enough to be repeated with confidence by something that has never met them.
Common questions
How should we track whether any of this is working?
The prompts are in the body of this post; the part people skip is the logging. Keep one sheet with the date, the exact prompt wording, and whether your firm appeared — and never change the wording, or the series stops being comparable. Answers vary from one run to the next even with identical prompts, so you are looking for a trend across several months rather than a reading on any one day. Two minutes a month, and it is the only record you will have of whether anything you changed made a difference.
Is this different work from ordinary SEO, or the same thing again?
Mostly the same foundation, with two additions and one genuine difference. The additions are agreement — the same facts about your firm appearing in more than one place — and answering questions at the top of a page instead of the bottom. The real difference is measurement. There is no position to track, so the only honest instrument is running a fixed set of prompts yourself and logging the results over several months, since any single answer varies run to run. A firm that is unfindable in conventional search will not be cited in an answer either, so this is not a channel you can switch to.
Should we be blocking AI crawlers rather than inviting them?
That is a legitimate question and the answer depends on which crawler. The systems that collect material to train models are identified separately from the ones that fetch pages to build live answers, and each can be allowed or refused on its own line. Blocking the training crawlers while permitting the retrieval crawlers is a defensible position. Blocking everything means accepting that you cannot be named, which for a firm selling to buyers who research before they call is an expensive form of privacy.
We already have a website. Why isn't that enough to get cited?
Because a website is a claim you make about yourself, and a system recommending you to a stranger is looking for someone else to agree. It also matters what kind of writing is on the site. A capabilities page presents; it does not answer. What gets lifted is a plain statement about a specific question, and most technical firm sites contain none, not because the expertise is missing but because nobody has ever been asked to write it down.
How long before any of this changes what an AI answer says?
Fixing access and correcting inconsistent business details can register within weeks, because those are reachability and agreement problems rather than reputation problems. Being named in response to a competitive open question takes longer — months, and it depends on accumulating outside mentions you do not fully control. Do the entity corrections first for a reason beyond speed: until the record resolves to one firm, later mentions attach to a company the system cannot quite identify as you.
Want this read on your business?
The Growth Signal Audit is a 30-minute review returned as a one-page written findings document — what is working, what is leaking, and the two or three things worth doing first. No cost, and no obligation to do anything with it.
Request a Growth Signal Audit