AI Only Recommends Firms It Can Recognize

Ask ChatGPT to recommend a structural engineering firm for a mid-rise timber building, or ask Perplexity which technical consultancy handles pressure vessel certification in Northern Europe, and the model has to do something specific: pick names out of everything it has read and decided is worth repeating. It cannot recommend a firm it has never encountered as a distinct, describable entity. It can only recommend a firm it can recognize.

That sentence sounds obvious until you notice what "recognize" actually requires from a services business, which has none of the shortcuts a product company gets.

A services firm has no datasheet to fall back on

A manufacturer that sells a physical product gives a model an easy trail: a spec sheet, a part number, reviews that mention the same model repeatedly. That redundancy is what makes a product recognizable across many independent pieces of text. A professional-services firm sells judgment, not a part number. There is no spec sheet for "we know how to get a chemical plant through an environmental permit review." The only trail such a firm can leave is written description: what it did, for whom, in what sector, solving what kind of problem, described specifically enough to be reused in an answer.

Most firms in the sector get this backwards. They assume that because their site describes their work, the description does the job. Usually it doesn't.

Generic capability language carries no reusable information

Open the "About" or "Services" page of almost any technical consultancy and you'll find some version of the same three sentences: a trusted partner, a track record of excellence, tailored solutions for every client. None of that is false, exactly. It's also not information. A model cannot restate "trusted partner" as an answer to "who handles X in market Y", because the phrase doesn't say what X is or which Y the firm has actually worked in.

What a model can restate is a sentence that names a service line, an industry, a region, or a problem type in specific enough terms to be reused without the model having to invent the specifics itself. "We provide structural assessments for older reinforced concrete bridges in coastal environments" gives a model something to work with. "We deliver excellence in structural engineering" does not, no matter how true it is or how long the firm has actually been doing that kind of work.

The mechanism isn't mysterious once you say it plainly: a language model answers by drawing on patterns in text it has read, and the more specifically a piece of text names an entity alongside a concrete capability, the more likely that association gets reused. Vague language creates no pattern worth reusing, because it says nothing that distinguishes the firm from thousands of other pages using the same three sentences.

Reputation and web substance are not the same asset

Every technical sector has firms well known among peers and buyers through referral alone: the name that comes up when a project manager asks a colleague who handled something similar. That reputation is real, and it is often why the firm never invested in its own written content.

Imagine, purely as an illustration, a structural engineering firm with a strong regional reputation built on referrals, and a website that says little beyond "trusted partner" and "extensive experience." Now imagine a smaller, newer firm in the same specialty that has published detailed write-ups of specific completed projects, spoken at industry conferences with the talks summarized on its site, and been mentioned in a trade publication for an unusual technical solution it used. The second firm is less known by reputation, yet more likely to be the one an AI assistant names, because referral conversations never made it into text a model can read, while the second firm's actual work did.

That asymmetry doesn't resolve itself. Being well regarded among people who already know you is a different asset from being describable to a system that has only ever read what's public.

What "recognizable substance" looks like for a services business

For a professional-services firm, that material usually comes from a few places: service lines described in specific terms rather than category language, named industries or regulatory regimes the firm actually works within, case studies on the firm's own site that describe a real completed project with enough specificity to be distinguishable from a generic one, technical articles published under the firm's own name, conference talks that get written up rather than only recorded, and mentions in trade publications or professional-body directories that describe what the firm does rather than just listing its name.

None of this requires marketing polish. A case study that reads like a dry internal project memo is more useful to a model than one rewritten by a copywriter into paragraphs of adjectives with the technical substance removed. The version that survives being folded into an AI-generated answer is the one with names, sectors, and problem descriptions left intact.

This is not something we fix by "optimizing" your content

We want to be precise about what PSentry actually does here, because this is exactly the kind of gap where it would be easy to overclaim. We don't rewrite your service pages, don't submit anything to a model, and don't influence what ChatGPT, Claude, Gemini or Perplexity decide to say about your firm. What we do is run your prompt set, the kinds of questions a buyer would actually ask, across all four platforms, and report which firms get named, whether you're one of them, and which competitors are named in your place.

For a firm that believes its reputation is strong, that report can be the first honest signal that reputation and AI recognizability have quietly diverged. What you do with it, publish more specific case studies, name service lines more precisely, put technical writing back online, is a decision for the firm. Measurement and correction are different jobs, and we only do the first one.

The multilingual version of the same problem

An engineering or consultancy firm operating across several export markets usually translates its marketing pages: the homepage, the services overview, the contact form. It rarely translates the technical writing, because that is expensive to do well and nobody budgets for it. The result is a firm that looks describable in its home language and close to invisible in every other language it operates in, not for lack of a track record in those markets, but because the record was never written down in that language.

A model answering a question in German about pressure-vessel consultancies draws on the German-language web. If your case studies and technical articles exist only in English, none of that substance is available to answer the German question, no matter how many projects you've delivered in German-speaking markets. This is a per-language problem, not a one-time translation task: the gap is rarely the same size in every market.

A check you can run yourself

Open your own services pages in the language of a market you operate in, and read them as if you were a model trying to answer "who handles X in this country." Count how many sentences name a specific service, sector, or completed project, and how many are interchangeable with a competitor's page if you swapped the logo. If the second group dominates, that page currently gives an AI assistant nothing to reuse, however good the underlying work is. Then repeat the exercise for a market where your reputation is weaker: the content gap and the reputation gap rarely line up the way you'd expect.

Frequently Asked Questions

Does a firm need a blog to be recognizable to AI assistants?

Not a blog specifically. What matters is that specific, checkable descriptions of the firm's actual work exist somewhere public: on its own site, in a trade publication, a professional directory, or a conference program. Format matters less than specificity.

Can a firm with no published content still get named by an AI?

It can, if other people have written about it specifically enough, for example detailed press coverage or a documented industry award. But relying on someone else to describe your work is a weaker position than doing it yourself, since you control your own site and nobody else's.

Is this the same issue as a brand not existing as an entity?

Related but narrower. Entity recognition is about whether a model can identify your organization as a distinct thing at all. This is about something more specific to services firms: even once a firm is recognized as an entity, generic capability language gives a model nothing concrete to recommend it for.

Should we rewrite our existing case studies?

That's a content decision for the firm, not something we make for you. What we can tell you is whether your firm gets named for the questions your buyers ask, in which markets it doesn't, and which competitors are named instead, so you can judge whether a rewrite is worth prioritizing.

Do all four AI platforms behave the same way here?

No. Perplexity retrieves and cites live sources when it answers, so freshly published content can surface sooner. ChatGPT, Claude, and Gemini lean more on what they absorbed during training, so new material can take longer to shift their answers, and results can vary between the two kinds of platforms for the same firm.

How often should a firm check this?

We scan twice a month, which fits how slowly firms typically change their published content and how slowly models absorb new material. Checking more often than the content itself changes doesn't produce more useful information.