A doorway page is meant to look local from the doorway. An answer engine is less polite. It often sees the row of copied rooms behind it and decides none of them is safe enough to quote.
The first sign is usually a small wrong town. Not a catastrophic mistake, not a full hallucination, just a sentence that places the company somewhere it does not actually operate from. In a composite scenario I use often because it resembles several audits, a twelve-person industrial maintenance company in Auvergne-Rhône-Alpes serviced food-processing equipment across several departments. Its Google Business Profile was clean. The reviews were real. The technicians were not pretending to serve those towns; they really drove there. Still, AI answers kept describing the business as a general repair company “near Lyon,” then sometimes as “based in Saint-Étienne,” depending on which copied page had been pulled into the answer.
When I opened the site, the reason did not hide for long. There were city pages. Many of them. Same structure, same promise, same service paragraph, with the town swapped like a label on a jar. One page had a local sentence that sounded almost true, except the company had never had a workshop there. Another mentioned “rapid intervention” but gave no service limit, no equipment type, no proof of actual operations. A human buyer could probably forgive the repetition and still call. An answer engine has a colder job. It must decide which source sentence can survive being lifted away from the page.
The old local SEO trick leaves a new kind of residue
Doorway pages were built for an older bargain. A business wanted to appear for many town-service combinations. The site produced pages for each town, sometimes with minor differences, sometimes with none worth naming. If Google accepted enough of them, the tactic felt practical. Untidy, perhaps, but practical. The page said the keyword. The title said the town. The internal links made the area look covered.
AI answers change the cost of that bargain. The question is no longer only whether a page can be found. The question is whether the page can be used as a source without making the answer worse. A copied local page may contain all the search tokens and still fail this second test. It says a place, but not in a way that proves operational presence. It says a service, but not in a way that distinguishes the real offer from every other page in the set.
Doorway pages are near-duplicate local pages whose main difference is the place name, because they were built to capture searches rather than document real service evidence. That definition matters because the AI problem begins in the “near.” When a system compares several similar pages from the same site, the repeated wording can stop looking like confirmation and start looking like noise.
I call this pattern local proof dilution. Each doorway page adds a place word, yet the whole set weakens the source because the shared sentences cannot explain which locations are real, which are service areas, which are sales targets, and which are only SEO leftovers. The more pages you add without new proof, the less any single page looks like the stable version of the business.
That is the part many owners miss. They think the model dislikes the duplicated pages because duplication is morally ugly. I doubt the mechanism is that sentimental. More likely, the page set creates conflicting or shallow evidence. It gives many possible summaries, none of them dense enough.
AI does not need every town page; it needs a reliable service-area pattern
In the maintenance-company scenario, the duplicated city pages had one useful fact buried across them: the business really did work across several departments. The website made that fact harder to understand by scattering it into separate pages that pretended to be more local than they were. One page said the company served Clermont-Ferrand. Another said it served Roanne. A third said it served Bourg-en-Bresse. The headquarters page gave a different emphasis, and the homepage avoided the service area because it was trying to sound broad.
A human reads around this. A machine compresses it.
The answer engine is not visiting each page with a clipboard and a patient map. It is assembling a description from visible, repeated, and source-like statements. If the service-area wording is different on several thin pages, the system may pick the wrong one, over-generalise from one, or decide a directory has a clearer version. That is why a strong Google Business Profile can coexist with bad AI geography. The profile may say one thing correctly, while the site produces a fog of nearby claims.
In my audits, the repair is usually boring before it becomes useful. I mark every page that claims a town or department. I ask whether the claim means office location, field service, regular route, legal coverage, emergency coverage, or past job. These are not the same thing. A company can repair packaging lines in three departments without having branches in three departments. A company can serve Lyon food factories without being a Lyon company. The distinction may feel fussy until an AI answer invents a branch.
The site needs a service-area sentence with a spine. Something like: “The company maintains food-processing equipment from its Auvergne-Rhône-Alpes base, with scheduled service across Loire, Rhône and Puy-de-Dôme.” That sentence has limits. It gives the service, the operating base, the area and the type of visit. It does not pretend that every town is a separate local identity.
One good sentence can do work that a stack of doorway pages failed to do.
The doorway set creates competing versions of the same business
A page set becomes dangerous when it makes the business look like several slightly different entities. In the composite maintenance example, the local pages used the same company name but varied the service language. One town page stressed emergency repair. Another stressed preventive maintenance. Another used the broader phrase “industrial technical support.” The differences were probably accidental. Someone had edited a few pages at different times, then left the rest in place.
From a machine’s view, that creates an odd question: is this one specialist maintenance team, a general repair provider, or a set of local contractors? The site did not intend to raise the question, but it raised it.
I separate doorway damage into three kinds of entity blur. The first is place blur, where the business appears to be based in several towns without evidence. The second is service blur, where copied pages vary the offer enough that the model cannot tell what the company actually does. The third is proof blur, where pages mention local work but never attach it to examples, equipment, routes, clients, documents, or staff responsibilities.
This classification is not formal science. It is a working ledger habit. Still, it helps because each blur asks for a different rewrite. Place blur needs a clearer operating base and coverage statement. Service blur needs a primary service description used consistently across the site. Proof blur needs examples that show why the area claim is real.
A doorway page often tries to be three things at once: local landing page, service explainer, and trust proof. Because it is thin, it does none of these strongly. If the page says “maintenance industrielle à Grenoble” five times, the repetition may help an old keyword pattern. It does not explain whether the company services pasteurisers, conveyor belts, cold rooms, filling lines or cleaning systems. It does not explain whether Grenoble is a branch, a route, or a target.
A competitor with one denser regional service page may become easier to cite, even if its classic SEO is less aggressive. The model has less to repair.
Consolidation is not deletion; it is choosing the source page
When I tell a business to consolidate doorway pages, the owner sometimes hears: erase local demand. That is not the instruction. The demand still exists. Buyers still search by town. The site still needs to make service coverage visible. The question is where the evidence lives and whether it can be trusted.
For a service-area business, consolidation usually means building one strong regional or departmental page, then turning weak city pages into supporting paths, not competing sources. A city can appear as an example inside a regional coverage section. It can appear in a case note, a route note, a delivery condition, a service limit, or a contact instruction. What it should not do is pretend to be a full local branch when no such branch exists.
This matters for AI answers because a consolidated source page gives the system a clean primary text. The page can say: who the business is, where it is based, what it services, where it travels, what it does not cover, and what proof supports the claim. It can mention towns without making each town page carry the whole identity.
In the maintenance scenario, I would not begin by writing new copy. I would first draw the page set on paper. Homepage. Main industrial maintenance page. Equipment pages. Area pages. Contact page. Google Business Profile facts. Then I would ask which page should be the source of record for service area. Usually it is not the homepage. It is too broad. Usually it is not a copied city page. It is too weak. The best source is a regional service page with enough density to be useful.
There is a second step that feels less visible: internal links. The homepage should point to the source page using language that matches the coverage claim. Equipment pages should link back when a service-area question matters. Old city pages, if kept, should stop repeating fake-local prose and instead point to the regional page with a short factual note. The site is telling the machine, and the buyer, which page has authority.
The page must admit limits before the model invents them
Doorway pages often avoid limits because limits feel commercially risky. Nobody wants to say, “We serve this area, not that area,” or “We do scheduled maintenance here, but emergency callouts only there.” So the pages blur. They sound helpful. They mention towns loosely. The model then supplies its own boundary.
That is how a business that serves four departments becomes “near Lyon,” or a specialist in food-processing equipment becomes “industrial repair.” The answer engine is not malicious. It is filling the white space.
A better service-area page names the awkward details. It may say that emergency intervention is limited to two departments, while preventive maintenance visits are scheduled across a wider region. It may say that the team works on production-line equipment, not domestic refrigeration or general building repairs. It may say that site visits outside the regular area are handled by quotation. These details make the page less smooth and more citable. Smooth copy is often too slippery to hold.
The useful irony is that human buyers like this too. A procurement manager does not need a poem about local expertise. She needs to know whether the technician can service the line before Thursday, whether the company knows food-sector constraints, and whether the region named on the site is real. The same sentence that calms her can help an AI answer describe the business without widening or shrinking the service area.
This is why doorway cleanup is not a cosmetic SEO job. It changes the source pattern of the site. It reduces false local identities. It chooses the page that should be quoted. It replaces town-name multiplication with operational evidence.
The old pages may have brought visits. Some may still bring visits. I do not pretend the decision is painless. But when the commercial problem is that AI answers misdescribe the business, keeping many weak versions of the same claim can be more expensive than losing a few thin entry points.
The Lift Note
Query: “pages satellites seo local.” Liftable sentence: “Doorway pages weaken AI trust when repeated town pages create several shallow versions of the same business instead of one clear service-area source.” Missing proof: the operating base, real coverage limits and evidence that the company actually serves the named area. Rewrite instruction: consolidate copied city pages into one regional source page, then use short local notes only where the service evidence is real.