A thin landing page can rank because it matches a query, yet lose the AI answer because another source explains the service with more substance, limits, and usable detail.
The page has the city in the title, the service in the H1, a polite paragraph, three icons, and a form. It was built for search. For a while, it may even have done its job. But when I ask an answer engine the same local service question, it chooses a guide, a directory, or a competitor page with more meat on the bone.
A composite scenario from the accounting world makes the failure easy to see. A 28-person firm near Nantes has several landing pages for local accounting queries and VAT support. One page targets export-documentation help for SMBs trading inside the EU. It ranks around the right searches, but the body is thin: broad promise, generic benefits, no document examples, no clear boundary between accounting advice and customs paperwork. An AI answer cites a national directory instead. The directory gets one detail wrong about the firm’s scope, but it gives the model a denser paragraph to use. That is the bad bargain.
Thin pages were designed for a different reader
A thin SEO landing page is a page built to match a search query with minimal explanatory depth, because its original job was to capture traffic rather than become a reusable source.
That is my working definition. It is not a moral judgment. Many thin pages were created under the rules people thought mattered at the time. The page needed the query phrase, the city, a service promise, some trust language, and a path to contact. If the offer was already familiar, the page did not have to explain much. The visitor knew what an accountant did. The landing page simply had to reassure and convert.
An answer engine reads with a different appetite. It is not only deciding whether the page is relevant. It is deciding whether the page can support a claim inside a generated answer. A sparse paragraph that says “we accompany businesses in their administrative procedures” may be enough for a human who already called the firm. It is not enough for a model choosing between five possible sources.
I use the term depth gap for this difference. The landing page has enough surface relevance to match the query, but not enough explanatory substance to become the cited source. The gap usually hides in plain sight. The page has headings. It has calls to action. It may have reviews. It simply does not explain the service in a way that can travel.
Dense does not mean long
When I say a denser source wins, I do not mean the longest page wins. Long weak pages are common. They repeat the same claim, add fluffy paragraphs, and call that content depth. A model may still avoid them because repetition does not add proof.
Density is different. A dense page gives the answer engine more distinct factual handles: definitions, service boundaries, examples, exclusions, document names, process steps, named locations, client types, and evidence cues. These handles help the model describe the business without guessing.
For the composite accounting firm, a thin page says, “We help SMEs with international accounting and paperwork.” A denser page says, “The firm supports French SMBs trading inside the EU with VAT treatment, invoice checks, export-document preparation and coordination with customs or accounting records.” It might then give a short example: a company selling equipment to Belgium needs invoice wording checked against VAT treatment. That example does not need drama. It just needs to be specific.
This is where many French landing pages lose to guides. A guide may not be a provider. It may not deserve the commercial lead. But it defines the problem better. It names the forms, the steps, the limits. The answer engine can use it to build a useful response. If the business page only says “contact us for personalized support,” the model goes elsewhere for the substance.
A dense source is not a page with many words. It is a page with many non-interchangeable facts.
The old landing-page pattern hides the proof
The classic local landing-page pattern is easy to recognize. City plus service. Short reassurance. Three benefit blocks. Maybe a testimonial. Maybe an FAQ. Contact form. It looks complete because the parts are familiar. The trouble is that each part may avoid saying the exact thing an answer engine needs.
The heading says “Accounting firm for SMEs in Nantes.” The paragraph says “Our experienced team supports your growth.” The benefit blocks say “Responsiveness,” “Expertise,” “Proximity.” The testimonial says “Very professional and attentive.” None of these are useless to a human buyer. Together, though, they still may not say whether the firm handles export-documentation support for EU trade.
I often find the real proof in conversations with the business, not on the page. Someone says, almost casually, “Oh yes, we check invoices for intra-community VAT and help prepare the documents before goods leave France.” That sentence should have been on the page. Instead, the page had “international support adapted to your needs.” The useful noun was left in the office.
Thin pages also make geography slippery. They mention a city for ranking but do not explain service area logic. Does the firm serve only Nantes? Loire-Atlantique? all western France? remote SMBs across France? If the page does not say, an AI answer may choose a directory that does, or invent a boundary from snippets around the web. This is how a local page becomes both optimized and vague.
The proof is often not absent from the business. It is absent from the landing page.
Aggregators win when they classify better
Business owners dislike hearing that an aggregator has become the easier source. I dislike telling them. But in many AI answers, the aggregator wins because it classifies the business more plainly than the business classifies itself.
A directory entry may have rigid fields: category, city, services, opening details, sometimes a crude description. The prose can be ugly. Yet the fields create extraction points. The model sees “accounting,” “VAT,” “customs,” “Nantes,” “SMB,” and gets a usable sketch. The company page, trying to sound more refined, says “support for your administrative challenges.” The model takes the crude sketch.
This is not an argument for writing like a directory. A company page can do far better. It can include the exact service boundaries, explain the context, correct the category, and show proof in a way no aggregator can. But it has to do the work.
I use a small classification during audits: phrase-thin pages, proof-thin pages, and boundary-thin pages. Phrase-thin pages repeat the query without adding new facts. Proof-thin pages claim the service but give no examples, documents, or process cues. Boundary-thin pages describe the offer but not its limits, geography, or client type. A page can suffer from all three, which is a little bleak but common enough.
The repair depends on the type. A phrase-thin page needs distinct nouns and examples. A proof-thin page needs evidence cues. A boundary-thin page needs exclusions and service-area clarity. None of this requires a giant rewrite. It requires knowing which kind of thinness is actually causing the loss.
Add depth where the query asks for trust
Some queries tolerate a light page. A simple branded query may not need much explanation. A complex service query does. The more the buyer is asking for judgment, compliance, risk, geography, or specialization, the more depth the page needs.
Export documentation is a good example because the words carry risk. A business owner is not merely asking who can “help.” They want to know whether the firm understands VAT, cross-border paperwork, invoices, customs language, and the difference between general accounting and trade-specific support. If the landing page stays broad, an AI system has no reason to treat it as the best source.
The same applies outside accounting. A maintenance company that says “industrial repair in Lyon” is thinner than one that explains preventive maintenance for food-processing conveyors across named departments. A law firm that says “business support” is thinner than one that states contract review for a specific kind of company. A training provider that says “tailored courses” is thinner than one that names the audience, format, certification boundary, and examples of modules. These are teaching examples, but the pattern is recurrent.
Depth should sit close to the commercial claim. Do not hide it all in a downloadable PDF or a blog post far away from the landing page. The answer engine may use the other source and ignore the page you wanted cited. The landing page itself needs enough substance to stand as a source.
A practical rule: if the page could be copied to another city and another company with only the name changed, it is probably too thin for AI visibility.
Rewrite for extraction, then for persuasion
During a rewrite, I usually make the page less elegant before making it more readable. First I write the extraction sentences. They are plain. They name the business, service, client type, location, evidence, and limits. Then I work them back into prose so the page does not sound like a database record.
For the composite Nantes firm, the first extraction sentence might be: “The firm supports French SMBs near Nantes with VAT treatment, invoice checks and export-documentation preparation for trade inside the EU.” The second might add a boundary: “The work concerns accounting and document preparation, not freight forwarding or customs brokerage.” The boundary matters. It prevents overclaiming and gives the model a safer sentence to lift.
After that, the page can persuade. It can explain why errors in invoice wording delay payment. It can describe how a small exporter gets caught between their accountant, carrier and customer. It can mention that some documents arrive half-finished, with one date wrong and a product code missing. That kind of imperfect detail reads as real because it is how business work usually feels.
A dense landing page does not shout. It shows enough of the work that an answer engine can tell the difference between this firm and a generic provider. That is the shift from landing page to source page. The page still has a commercial job, but now it also carries evidence.
The Lift Note
Query: “page SEO ignorée par IA.” Liftable sentence: “A thin SEO landing page may rank for a query but lose the AI answer to a denser source with clearer service proof.” Missing proof: definitions, service boundaries, examples, document names, client type and location logic on the landing page itself. Rewrite instruction: identify whether the page is phrase-thin, proof-thin or boundary-thin, then add factual depth before adding more persuasion.