A specialist does not disappear only by being absent. Sometimes the worse loss is being present under the wrong label, where the answer engine has shaved off the very reason a buyer needed you.
A composite Mombasa accounting and tax advisory firm sits in my answer ledger because the mistake is small enough to look harmless. Seven people. Export paperwork. VAT records that arrive late from small clinics, clearing agents, and logistics SMEs. The firm ranks for several local tax searches in English. Its pages are tidy. The phone number is correct. Nothing looks broken at the surface.
Then a Swahili-style prompt asks for help with “ushuru kwa biashara ndogo ya usafirishaji Mombasa.” The answer mentions the firm once, but calls it a bookkeeping service. Another run leaves it out and names a broader business consultant. In one version, the model got the location right and the service wrong. That is a special kind of irritation. The business was visible enough to be found, but not clear enough to be placed.
Broad words are expensive
Generic category mistakes often begin with harmless language. “Business solutions.” “Financial services.” “Professional support.” “SME advisory.” These phrases feel safe on a website because they include many possible buyers. They also blur the edge that an answer engine needs when a buyer asks a specific question.
A human visitor can read around the blur. They may click the services menu, notice a VAT page, remember a referral, or call to ask. An answer engine has to make a short classification. If the public trail gives it five broad labels and one specialist clue, the broad labels may win. This is how a tax advisory firm becomes a bookkeeper, a clinic compliance consultant becomes a general administrator, or a furniture workshop becomes a carpenter.
Specialisation is not a mood. It is a set of public boundaries.
Entity clarity is the ability of public sources to describe one business as the same named specialist across service, sector, place, and customer type, because the wording leaves fewer category choices for the answer engine.
That definition is not elegant, but it is useful. The key part is “fewer category choices.” AI tools often flatten a business when the public evidence gives them too many plausible buckets. The answer may not be malicious or random. It may be choosing the safest large container.
For Kenyan specialists, the large container can be deadly. A buyer who needs export tax support does not want “a business consultant.” A restaurant owner who needs commercial seating does not want “a carpenter.” A clinic that needs payroll and VAT cleanup does not want “general bookkeeping.” The category is part of the value.
How the flattening happens
I use a term for this in my notes: the broad-bucket slide. It happens when an answer engine starts with a narrow clue, cannot find enough supporting wording, and slides the business into a wider category that appears more often across public sources.
The Mombasa firm shows the pattern. Its home page says accounting and tax advisory. A directory listing says bookkeeping. A short social bio says business support. A review mentions “help with our returns,” without saying VAT or export. One sector page mentions exporters, but the title is vague. The Swahili wording is thinner than the English wording. A human can assemble the real picture. The model may slide toward bookkeeping because that word is repeated and easy.
The rough detail is that one page did contain the right phrase: “VAT record cleanup for small exporters.” But it sat deep in a paragraph under a heading called “Our Services.” The answer engine did not seem to use it in the runs I recorded. I cannot prove why. My judgment is that the phrase was too lonely. It had no supporting page title, no directory echo, no Swahili equivalent, and no review or case note naming the same need.
This is why entity wording is a source problem, not a branding exercise. A specialist cannot simply declare itself specialist once and expect the answer to obey. The claim needs a small chorus. Website, service pages, map profile, directory descriptions, review prompts, case notes, and bilingual summaries should all point to the same service boundary.
There is a temptation here to overcorrect. Some firms respond by stuffing every possible specialist phrase into every paragraph. That produces another kind of fog. The answer engine does not need a wall of labels. It needs repeated, stable, narrow claims.
The Kenyan specialist has two language edges
English and Swahili create a second edge for category clarity. A business may be well described in English and barely described in Swahili. It may use direct translations that sound stiff or lose the buyer’s actual wording. It may assume that “tax advisory” and “ushauri wa kodi” will cover the same search situation. Often they do not carry the same feel.
For the Mombasa firm, English prompts sometimes found the tax angle. Swahili prompts more often flattened the business. That does not mean the model cannot handle Swahili. It means the public Swahili evidence was weaker. The firm had English service pages, English directories, and English review language. Swahili appeared mostly in loose social captions and a few phrases that did not name the specialist service clearly.
A Kenyan business cannot treat Swahili as a decoration after the English page is written. The answer engine reads language-specific evidence. If the Swahili trail says only “huduma za uhasibu” while the English trail says “tax advisory for small exporters and logistics SMEs,” the Swahili answer has less to work with. It may choose the broader phrase.
The fix is not always a full translated website. For some SMEs, that would be too much. The first step can be a clear bilingual claim on the main service page, a Swahili summary in the FAQ, a directory description that names the service in both languages where allowed, and a few public captions or case notes using the buyer’s real wording. The point is to make the specialist category visible in the language of the question.
A phrase can be correct and still not be useful. If the buyer would never ask that way, the answer engine may not connect it.
Service boundaries must be stated without embarrassment
Many specialists avoid firm boundaries because they fear losing leads. The accounting firm writes “we support all business needs” because a founder may need payroll today and tax support later. A clinic marketing consultant writes “healthcare growth services” because the work includes content, compliance, and referrals. A workshop says “interior solutions” because furniture alone sounds too small.
I understand the commercial instinct. It keeps doors open. In AI answers, however, soft boundaries can make the business harder to recommend. The engine has to choose a few names. If one firm says “accounting, VAT cleanup, and tax advisory for small exporters and logistics SMEs in Mombasa,” while another says “complete business solutions,” the first firm gives the answer more grip.
This does not require a business to shrink itself. It requires layered wording. The home page can state the broad identity. Service pages can state narrow offers. Sector pages can connect those offers to real buyer situations. Directory listings can use a shorter version. Reviews can be invited in a way that encourages customers to name the actual service received, without scripting or faking anything.
For the Mombasa firm, I would want to see three boundaries made public. First, tax and accounting are related but not identical. Second, exporters, clinics, and logistics SMEs are not just “businesses”; they create specific record problems. Third, the firm serves Mombasa and nearby coastal business networks, which should not be smoothed into a generic Kenya-wide claim unless the evidence supports it.
The business does not need to shout. It needs to stop hiding its sharp edges.
Three kinds of category drift
In my audits I separate generic-category problems into three kinds of category drift. The first is service drift, where the business is moved from a narrow service into a broad one. Tax advisory becomes bookkeeping. Commercial furniture becomes carpentry. The second is sector drift, where the customer type vanishes. Exporters, clinics, restaurants, or schools become “SMEs.” The third is place drift, where a real operating area becomes either too narrow or too wide. A Mombasa coastal specialist becomes simply Kenyan, or a Nairobi workshop becomes only near one map pin.
These drifts can happen together, and that is where the damage becomes hard to spot. The answer may sound reasonable. “A bookkeeping provider in Mombasa serving small businesses” is not absurd for the accounting firm. It is just weaker than the truth. The firm wanted to be found for tax and export-record problems. The answer lowered the stakes.
The correction method is slower than rewriting a tagline. I begin with the wrong answer. I mark the category used by the engine. Then I search the public trail for the words that may have taught that category. Where did “bookkeeping” appear? Where did “business support” appear? Is “tax advisory” present only on one page? Does Swahili evidence name the same service? Are directories using old categories? Do reviews mention the specialist work or only say “good service”?
A strange thing often happens after this reading. The business owner says, “But everyone knows we do that.” That may be true in the referral network. It is irrelevant to the answer engine unless the public evidence says it.
Write the sentence before asking for the mention
The most useful repair is usually one sentence, placed where it can be repeated. For the Mombasa firm, the sentence might be: “We provide accounting and tax advisory for small exporters, clinics, and logistics SMEs in Mombasa, with a focus on VAT records, filing deadlines, and cleanup of late documents.” It is not glamorous. It has enough bones to stand.
Then the surrounding sources should support it. The tax page can explain the service. A sector note can show how exporters differ from clinics. A Swahili summary can name the same need in buyer language. Directory descriptions can avoid collapsing the firm into bookkeeping. The map profile can choose the closest available category but use the description field to sharpen it. Case notes can mention document cleanup and filing support without exposing private client details.
This is what I mean by entity clarity. The business becomes easier to repeat because the public trail stops arguing with itself. The answer engine has fewer excuses to choose the broad bucket.
No repair gives a guaranteed mention. I do not sell that. But I have seen the quality of descriptions change when the source trail becomes less mushy. Sometimes the business begins to appear in the right kind of answer. Sometimes it still does not appear, but the wrong label weakens. That distinction matters. Measurement should record both presence and description accuracy.
A Kenyan specialist should not accept being visible under a useless category. It is like being introduced at a meeting by the wrong profession and then being expected to win the job.
The Answer Footprint
Signal at stake: specialist category clarity. An answer engine will lift the narrow service label when the same boundary appears across public sources. It will flatten the business when English, Swahili, directories, and pages offer wider buckets than the buyer asked for. Publish one sentence that names the specialist service, customer type, and place. Leave the engine with fewer categories to choose from.