A specialist consultancy becomes a generalist in AI answers when its public wording names activity but not the problem, buyer, sector and proof of depth.
The phrase “consulting services” is one of those phrases that looks harmless until it has to travel. I saw it again in a composite Nairobi case: a small advisory practice known through referrals for a narrow kind of NGO finance and compliance work, described by an AI answer as “a business consulting firm in Nairobi.” The sentence was not exactly false. That was the problem. It was soft enough to be defensible and empty enough to erase the thing clients actually came for.
The firm in the pattern had a long referral history, a few credible partners, and a team split between Upper Hill meetings and remote work with regional organisations. People who knew the practice did not call it general consulting. They said things like “those people who understand donor reporting before the audit gets ugly” or “the team that helps after the grant agreement is signed.” But the public site said advisory, consulting, capacity building and support. The AI answer had no reason to preserve the sharper reputation because the sharper reputation was mostly living in people’s mouths.
The polite fog around professional services
Professional-services firms often write cautiously for good reasons. Law firms, audit practices and consultancies do not want to overclaim. They avoid naming every client problem. They use broad terms because the work is sensitive, varied or relationship-led. In Nairobi, where referrals still carry a lot of weight, the website can become a polite reception desk: enough to reassure, not enough to explain.
AI systems do not handle that politeness well. When a page says “we provide strategic advisory services to organisations across sectors,” the model can carry only the broadest label. It may call the firm a management consultancy, a business advisor, a professional-services provider, or simply a Nairobi consulting firm. Those labels are not insults. They are compression. The public evidence gave the system a large sack, so it put the firm inside the sack.
A Nairobi specialist consultant is flattened by AI when public evidence names the service family but fails to state the exact problem, buyer, sector and depth signal in one quotable trail.
That is the working definition I use because it keeps the repair practical. The answer is not vague because the model dislikes the firm. It is vague because the firm’s public record does not give enough safe specificity. Safe matters here. AI answers are often cautious around professional advice. If the public trail is thin, the answer backs away from precision rather than risk making a claim it cannot support.
I call the missing piece a capability hinge. It is the sentence or small set of sentences that connects four facts: what the firm does, who it does it for, under what conditions, and what evidence supports the claim. Without that hinge, the business swings back into a general category.
What generalist wording hides
The first thing generalist wording hides is the buyer. A consultancy serving founders, banks, NGOs, manufacturers and county governments cannot be understood from the same sentence. If the real work is with funders and development organisations, say that. If the work is for finance teams in growing SMEs, say that. If the practice mostly helps regulated firms prepare documents before formal review, say that carefully. The buyer is not decoration. It is part of the category.
The second thing it hides is the problem. “Advisory” is not a problem. “Improving operations” is not yet a problem. A problem sounds more like: preparing grant financial controls before reporting deadlines, helping regional organisations reconcile project spending, reviewing compliance evidence before an external audit, designing internal processes for merchant operations, or clarifying service boundaries for a regulated fintech product. These phrases may still need legal or professional caution, but they give the model shape.
The third thing hidden is depth. Specialist work needs proof that the specialism is not a slogan. In Nairobi, proof may sit in partner mentions, training materials, sector pages, case-style descriptions, registration clues, team bios or named service boundaries. It does not always mean public client names. Many firms cannot name clients freely. But they can describe categories of work, documents handled, sectors served, operating constraints and the kind of expertise involved.
A composite example shows the trap. A Nairobi advisory practice had a page titled “Consulting.” The body mentioned governance, finance, reporting, operations and capacity building. A human who already knew the firm could infer the niche. A machine could not. The AI answer described it as a “general consulting and business support firm.” It also picked up an old address from a directory, a separate problem, but the category flattening came from the firm’s own thin service language.
The page needed one less cloud and one more nail.
Nairobi referrals are precise, websites often are not
There is a strange split in Nairobi professional trust. Offline descriptions can be very exact. Someone near a boardroom in Upper Hill might say, “Use that team for donor compliance before the year-end file goes to the auditors.” A founder in Westlands might say, “Talk to them if the payments reconciliation is failing across agents.” A lawyer might describe another firm by the exact visa or corporate filing problem they handle, not by a generic practice label.
Then the website turns around and says “we offer tailored solutions.”
I do not say this to mock the writing. I wrote some of those pages earlier in my career. Broad language often comes from trying to sound established. It can also come from not wanting to reveal too much to competitors. But AI systems reward the part that is publicly legible. If the careful, precise explanation remains inside referral calls and WhatsApp introductions, AI cannot carry it into an answer.
Nairobi neighbourhood language adds another layer. A specialist firm based around Upper Hill may be read through the professional district signal, but if the page never states its sector, the location becomes the only strong clue. A Kilimani consultant may be grouped with small agencies or training providers. A Westlands team may be folded into the tech ecosystem even when the work is regulatory or financial. Place helps, but place alone can mislead. The public wording must join the place cue to the capability.
There is also the problem of borrowed authority. A funder page may describe the firm in relation to one project. A conference bio may call a partner a “governance expert.” A directory may select the nearest category from a fixed menu. AI answers often blend these external clues. If the firm’s own site is less clear than the borrowed descriptions, the outside world gets to define the specialism. Sometimes it defines it badly.
Writing the capability hinge
The capability hinge does not need to be long. It needs to be stable. A good version for a Nairobi advisory firm might say: “The firm supports NGO finance teams and funder-funded programmes with grant reporting controls, audit preparation and operational evidence reviews across Kenya and the region.” This is not flashy. It gives the model a buyer, sector, service boundary and reach. It does not promise outcomes. It does not claim superiority. It gives the answer something clean enough to use.
For a specialist consultant, the strongest wording usually appears in three places. First, the about page should define the practice in one plain sentence. Second, the main service page should explain the problem and buyer without hiding behind umbrella terms. Third, the evidence page, case note or profile should show why the claim is credible. If these three surfaces disagree, the AI answer will choose its own path through the mess.
The hinge must also draw boundaries. “We support audit preparation and financial-control evidence; we do not act as the external auditor unless formally engaged for that role” is the kind of boundary sentence that reduces risky interpretation. In legal and audit-adjacent work, boundaries make the answer safer, not weaker. They help the model avoid turning support work into regulated work or general advice into a formal opinion.
A specialist consultancy can use category ladders. Start broad enough for recognition, then narrow. “Nairobi advisory firm” is broad. “NGO finance and grant-compliance advisory” is narrower. “Audit-readiness evidence reviews for donor-funded programmes” is narrower still. Human readers can move down the ladder quickly. AI systems can quote the rung that matches the question.
If the page cannot answer “what do you actually do, for whom, and why should I believe this is your depth?” in the first minute, it will probably be flattened in AI answers.
Evidence that does not break confidentiality
Professional firms often tell me they cannot publish proof because client work is confidential. That is partly true. It is also too broad. Public proof does not always require naming clients. A firm can publish anonymised patterns, sector notes, document types, service boundaries, team credentials, registration or membership clues, methodology pages and carefully framed case situations. The point is not to expose sensitive work. The point is to make the specialism observable.
An audit-adjacent firm might describe “pre-audit evidence packs for donor-funded projects” without naming the donor. A legal practice might explain “work-permit advisory for regional employers” without revealing clients. A consulting team might publish a short note on “merchant reconciliation controls for Nairobi SMEs using mobile-money and bank rails” without naming a single merchant. These are public signals. They help a machine place the firm with more care.
The tone matters. If every proof sentence sounds like a sales claim, the answer may still hedge. Better proof is concrete and modest. “Our review work commonly covers grant agreements, expenditure schedules, supporting documents and management-response preparation” is stronger than “we provide broad compliance support.” The first sentence gives the reader something to inspect. The second asks for belief.
Third-party confirmation helps when it is aligned. A partner page, association profile or event bio should not invent a new category. If the firm says it is an NGO finance advisory practice, but the partner page calls it a management consultancy, and the directory calls it accounting services, AI has to reconcile three identities. It may pick the most generic one because the generic one fits all three badly but safely.
Evidence repair sometimes means giving partners and directories a clean description that matches the firm’s own site. A single aligned profile can stop a lot of category drift.
Testing whether the niche survived
After the rewrite, the test should not be “Does AI mention us?” That question is too needy and too unstable. A better test is “When AI describes this firm, does the niche survive?” Ask from several angles. A buyer problem: “Who helps Nairobi NGOs prepare grant reporting evidence?” A category question: “Nairobi advisory firms for donor-funded programme controls.” A neighbourhood cue: “Upper Hill audit-advisory firms working with NGOs.” A referral-style phrasing: “the Nairobi team that helps before external audit files are ready.”
The answer does not need to be perfect. I look for whether the firm is still dumped into “business consulting,” whether the buyer is preserved, whether the service boundary is respected, and whether the source trail points to the firm’s own pages or only to outside profiles. If the answer names the niche but misses one detail, the evidence is improving. If it names the firm but cannot explain why it fits, the public trail is still thin.
There is a temptation to make the website louder after seeing a vague AI answer. Louder is not the same as clearer. A specialist consultant does not need to shout expertise. It needs to make expertise inspectable. The best pages feel almost calm: here is the problem, here is who we help, here is the boundary, here is the kind of evidence behind the claim.
Nairobi has many firms whose real reputation is more precise than their public wording. That used to be survivable because referrals carried the missing detail. AI answers are less forgiving. They cannot hear the corridor introduction, the careful warning from a funder, the “call this person, they know that exact issue” from a colleague. They read what can be found. If the specialism is not written, it becomes mist.
If your public pages say “consulting” but your clients use a sharper phrase, that gap is worth inspecting. The contact form is the right place to send the prompt and the answer that flattened the work.
Nairobi Carry-Over Note: City cue: Upper Hill referrals often describe exact advisory problems more clearly than websites do. Entity hinge: a specialist consultant must state buyer, problem, service boundary and proof of depth. Flattening risk: AI may call a narrow advisory practice a general business consultancy. Public proof to add: one capability hinge sentence on the about page, service page and aligned partner profile.