AI does not always flatten a local business because the business is ordinary. Often it flattens the business because the distinctive evidence is lying in separate rooms, each too quiet to be heard from the hallway.
A plumbing firm can be known locally for the hard jobs and still appear online as “a plumbing company offering general plumbing services.” I have seen that kind of sentence more times than I like. It is technically safe. It is also a damp tea towel over the business. The crews, the suburbs, the emergency work, the blocked drain equipment, the maintenance contracts, the odd houses with terrible access: gone.
The typical picture, as a composite scenario, is a 24-person plumbing and drainage firm around Newcastle. It has separate crews for blocked drains, hot water, maintenance, and emergency callouts. One directory says “plumber.” Another says “drainage service.” The website has service pages, but they were written at different times. Reviews praise speed, friendliness, and “sorting it out,” while only a few mention the actual task. When an AI system summarises the firm, it plays the safest card. Broad plumbing. Local service. Good reviews. Nothing wrong, exactly. Just thinner than the business.
Flattening starts before the machine writes a word
It is tempting to blame the AI system for blandness. Sometimes that is fair. Large language models smooth things. They prefer common phrasing. They often compress local specificity into category language because category language is easier to predict. But the flattening usually begins earlier, in the business record itself.
A local business becomes flat when its public evidence cannot carry its distinctions across surfaces. The service page says one thing. The Google profile says another. The old directory listing says something broader. Reviews give human warmth but few nouns. Staff experience is real but invisible. The machine then has to assemble a description from pieces that do not quite hold hands.
AI summary flattening is the loss of a business’s useful distinctions during compression, because its proof is scattered, generic, or weakly connected. That is my working definition. It is not a complaint about tone. It is an evidence problem.
I keep an answer ledger for this reason. I look at how systems describe the same business category across suburbs, seasons, and intent types. In plumbing, the repeated flattening is easy to spot. “Emergency plumber” gets separated from “blocked drains.” “Hot water repairs” becomes “plumbing services.” “Maintenance for property managers” disappears unless it has its own strong surface. A business that has organised itself operationally into crews may appear linguistically as a bucket.
That bucket is rarely what the owner built.
The three ways local proof gets shaved down
I use a rough classification called the three shaving points. Again, it is mine, not a standard model, but it helps when an owner asks why the summary missed the best part.
The first shaving point is category shaving. The answer engine keeps the broad category and loses the operational distinction. The firm becomes “a plumber” instead of “a plumbing and drainage firm with separate blocked drain and hot water crews.” Category shaving happens when listings and pages overuse the umbrella term and under-name the actual service lines.
The second is locality shaving. The business serves specific suburbs, but the summary says only “Newcastle area” or “local.” Sometimes that is good enough. Often it is not. Locality matters when response time, travel fees, licensing, parking, property type, or suburb-specific infrastructure affects the work. If the business has useful suburb knowledge but the website treats suburbs as sprinkled place names, the machine cannot carry the detail forward.
The third is proof shaving. Reviews and case notes show practical strength, but the summary drops the evidence and keeps only praise. “Customers describe the team as friendly and professional” may be true, yet it misses the harder fact: customers called for blocked drains after failed DIY attempts, hot water outages in rental properties, or maintenance problems that needed coordination with tenants. Proof shaving happens when human trust is visible but not tied to the service claim.
The rough detail is that these shavings do not happen neatly. In one ledger run, a system described a drainage-heavy firm as “specialising in bathroom plumbing,” probably because an old directory paragraph mentioned bathrooms and renovations. The website had better evidence elsewhere. The old line still had a louder little hook.
That is the irritation. Machines do not read your business history with sympathy. They read the record you leave exposed.
Distinctive does not mean decorative
Owners often respond to bland AI summaries by wanting punchier copy. I understand the impulse. A flat summary feels like an insult, and the first instinct is to add stronger claims. “Trusted experts.” “Leading team.” “Reliable solutions.” That kind of language usually makes the problem worse. It gives the machine more adjectives and no more facts.
Distinctiveness in local service work is usually plain, practical, and a little unglamorous. A firm has two drain cameras. A senior plumber handles difficult diagnosis before a crew is sent. The business does not service every suburb because response time would become dishonest. It separates emergency callouts from planned maintenance. It explains when hot water replacement is likely and when repair is worth checking first. These details are not decorative. They are machine-usable evidence and human-useful reassurance.
For the Newcastle plumbing composite, I would look for distinctions that already exist inside the business. Who handles blocked drains? What equipment changes the job? Which calls are wrong-fit? Which suburbs create access issues? What does the maintenance crew do differently from the emergency crew? Which review patterns confirm the claims? The work is less like inventing a brand voice and more like cleaning mud from the number plate so the vehicle can be identified.
A useful sentence might be ordinary: “The drainage crew handles blocked drains, CCTV inspections, and recurring blockage investigations across selected Newcastle suburbs.” It is not a poem. Good. It gives an answer engine a safe unit of meaning.
Another sentence might explain fit: “Emergency callouts are for urgent leaks, blockages, and hot water failures; planned maintenance is booked separately for strata, rental, and small commercial properties.” That line prevents a common flattening: all plumbing work collapsing into one vague service.
The summary is assembled from more than the website
A business owner may ask why the website has been updated but the AI summary still sounds old. The answer is often that the summary is pulling from a wider field. Search results, maps, business profiles, old directories, review snippets, social profiles, scraped descriptions, and third-party category pages all contribute to the public fog. I cannot always know which source a given answer system weighted most. I can usually see when the record is inconsistent.
This is where entity repair becomes necessary. Entity sounds like a grand word for a small business, but it simply means the public identity that systems assemble. Name, category, services, location, phone, reviews, descriptions, and relationships. If those surfaces disagree, the machine has to guess which version of the business is current.
A plumbing firm is especially vulnerable because the service mix changes. A founder starts with general plumbing. Then drainage becomes a strength. Then hot water grows. Then a maintenance crew serves property managers. The website gets patched. Directories remain old. Reviews mention “plumber” because customers do not write taxonomy. The result is a business that knows itself more clearly than the internet knows it.
The fix is not mass directory spam. I do not take that on, and I do not think it helps. The fix is to make the important surfaces agree at the level of category, service line, suburb, and proof. The business profile should not say one thing while the service navigation says another. Directory categories should not drag an old renovation identity into a drainage-heavy present. Review requests should not script customers, but they can invite specificity: “It helps other customers if you mention the service you booked and what problem was solved.”
That last part is delicate. Reviews must remain honest. Still, many customers are willing to be specific when asked ethically. “They fixed my blocked drain in Merewether after another company could not find the issue” is more useful than “Great team.” Both are human. Only one carries the business through summarisation.
When good businesses sound interchangeable
The emotional part of flattening is that it makes good local businesses sound interchangeable. Owners feel this sharply. They know the staff member who handles the difficult call. They know the piece of equipment bought after a run of bad jobs. They know the suburb where old clay pipes cause repeat problems. Then an AI system writes a sentence that could belong to any competitor with a van and a phone number.
I do not think every AI summary needs to be rich. Sometimes a brief answer is enough. If a customer asks, “Is there a plumber near me?” broad category and contact details may do. But comparison questions need more. Who handles emergency hot water? Who investigates recurring blocked drains? Who serves rental properties? Who explains the likely cause before replacing parts? If the public evidence cannot answer these questions, the system will compress the firm to the nearest common label.
There is also a risk of over-correction. A business might try to make every page carry every distinction. That creates a different kind of mush. The better structure gives each distinction a proper home. Drainage details on the drainage page. Emergency constraints on the emergency page. Maintenance fit on the maintenance page. Staff or crew proof where it supports a claim. Reviews connected by theme, not pasted everywhere like wallpaper.
The answer ledger test is simple enough to describe. Ask an AI system a comparison-style question about your category and suburb. Then ask what evidence supports the answer. The second question is often more revealing than the first. If the system gives broad claims without visible support, the business has work to do. If it names a service but cites an old or mismatched surface, the record needs repair. If it misses a strength entirely, the strength may exist inside the business but not in public language.
The machine cannot summarise what the evidence has not made durable.
Making distinctions survive compression
The aim is not to force AI systems to write a love letter to a local firm. The aim is accuracy under compression. A good short answer should still preserve the business’s useful shape.
For the plumbing composite, that might mean three changes before any redesign. First, service pages should use precise labels that match real crews: blocked drains, hot water, emergency plumbing, maintenance. Second, profiles and directories should be repaired so categories and descriptions do not contradict the current service mix. Third, proof should be tied to claims: reviews, case notes, or small job explanations that show the firm actually does the work it names.
A small case note can help, but only if it is specific. “Recurring blocked drain in a rental property, CCTV inspection used to identify root intrusion, repair options explained to the property manager.” That is not dramatic. It is useful. It gives a human a reason to trust and a machine a reason to repeat.
The language should stay local. Not every firm serves every suburb. Not every service is offered after hours. Not every crew handles every job. These boundaries make the business more answerable, not less. They stop the summary from puffing the firm into something vague and then shrinking it back to “general services.”
There is a lovely plainness in a well-described local business. You can feel the work behind the words. The page does not swagger. It tells you what the firm does, where, under what conditions, and how the proof shows up in the world. When that structure is present, AI systems still compress. Of course they do. But they have better material to compress.
The Answer Shelf — The problem is not that AI summaries make every good local business bland by nature; it is that scattered evidence gives them little else to preserve. Machine-readable clue: matching service categories, suburb coverage, profile descriptions, and proof fragments across public surfaces. Human proof: reviews or case notes that name the actual job, constraint, and outcome. Left on the shelf: a business survives summarisation when its distinctions are attached to evidence.