If your site only has short product sheets, Claude dismembers you: it cuts your text into pieces, mixes them with those of ten competitors, and loses track of who you are. The result is that in the answers your brand disappears into a generic category blob. A long, structured piece of content, on the other hand, enters intact, with your identity untouched — and Claude uses it as a recognizable source. There is a precise way to build content that Claude treats as a whole unit, and it doesn't require rewriting everything from scratch.
Sommeliers use Claude Projects with the wine list uploaded. If your wines don’t have indexed tech sheets, you’ll never make it into their project — and they won’t recommend you to clients.
This is the symptom that reached me a few months ago from a winery in Sardinia selling Cannonau and Vermentino all over Italy. The site was there, the product sheets too, but when professional sommeliers queried Claude to build pairings or update their wine list, the winery was almost never cited. The competition was. The reason is structural and has to do with how Claude reads documents: in one block, whole, up to 200,000 tokens at a time.
Let me explain why on Claude long and complete content has a specific advantage, and what changes for your visibility in AI answers when you stop chopping content into pieces and start writing real guides.
What it means to have 200,000 tokens of context
An AI model reads your content in units called tokens (to understand how the text-to-token conversion works, I covered tokenization in AI engines in a previous article). The wider the context the model can hold in a single read, the more documents, rules and data it can keep together within the same answer.
Anthropic’s Claude can hold up to 200,000 tokens in a single window. Crudely translated: roughly 500 pages of text, or a dozen white papers, or the entire wine list of a restaurant with all the technical sheets of the producers.
In the world of applied research, the topic is documented by the very team that builds Claude. Anthropic’s work, “Introducing Contextual Retrieval” (2024), describes how traditional RAG systems are forced to break knowledge bases into chunks of a few hundred tokens, and how this fragmentation reduces the context available to the model, worsening the retrieval of relevant information.
From this follows a consequence the paper doesn’t discuss explicitly from an editorial angle, but which I apply every day in content work: when a user uploads a document directly into Claude — via Projects, Artifacts or conversation — those 200,000 tokens are read whole, without the chunking that instead affects content retrieved from the web via RAG. In operational terms: if your content is a well-structured 8,000-word guide, it enters Claude all at once. If your content is a 200-word product sheet pulled via retrieval, it will end up merged into a chunk with ten other similar sheets and will lose its identity.
Why this ties together all the rest of the work you’ve already done
This doesn’t replace the other rules of the AI game. It sits above them. If you’ve worked on E-E-A-T for AI and on author entity recognition, the long guide you put online also becomes the piece that best demonstrates real expertise. Nobody writes 8,000 sensible words on a topic without truly knowing it.
Structure matters too: a long guide that doesn’t respect the inverted pyramid is read in full by Claude, but the first section remains the one that weighs most in shaping the answer. Length is no alibi for disorder.
The blog with 400 articles of 600 words.
The case of the Sardinian winery — what changed in 4 months
Back to the winery in Alghero. The initial problem was this: the hotel and restaurant sommeliers who work on Sardinian wines (Cannonau di Sardegna DOC, Vermentino di Gallura DOCG) increasingly use Claude as a study assistant. They upload the venue’s wine list into their project, ask for pairings, update selections. If your winery has no dense material that fits into the project, you disappear from the professional radar.
The intervention we tested as an editorial standard was this:
- A complete technical sheet for every wine, 1,500-2,500 words: vinification, specific terroir (slopes, altitude, soils), microclimate, tasted vintages with personal notes, regional and non-regional food pairings, serving temperature, evolution in the bottle, comparison with neighboring wines of the same DOC.
- A pillar guide per grape variety of 8,000-10,000 words: all the winery’s knowledge of Cannonau or Vermentino, from ampelographic history to the differences between sub-zones (Mamoiada vs Jerzu vs Oliena for Cannonau), down to the tasting vocabulary.
- Free downloadable PDFs of the technical sheets, well named, with correct metadata.
After four months, in repeated tests on Claude — asking for “Cannonau pairing with Sardinian suckling pig”, “difference in Vermentino di Gallura between Monti and Tempio”, “reliable Cannonau producers in Alghero for the restaurant trade” — the winery entered the answers consistently. An indicative test, not a controlled study: a sample of about twenty queries on Claude, with and without uploading real wine lists. The pattern, however, was clear and repeatable.
The point isn’t “winery X improved”. The point is that dense, long content overtook the competitors’ short content at the very moment professionals began using Claude as a daily work tool.
For every main product/service create a dense technical sheet of 1,500+ words, with a publicly downloadable PDF.
What you can test in 15 minutes for your industry
Open Claude (the free or paid version, claude.ai is fine too). Do two simple things:
- Ask a realistic query from your industry, like “best producers of [your product] in [your region] for [type of B2B client]”. See if you show up, and how many words it dedicates to you.
- Upload a PDF of your product/service sheet and ask Claude to compare it with three competitors you name. If Claude already knows enough about your competitors to make the comparison, but not about you, you have a public information density problem.
It’s an entry-level check, it doesn’t replace a real analysis with professional tools and observation logs over several months. But it immediately gives you an idea of where you stand and how you’re positioned.
What are the most frequent mistakes?
The 150-word product sheets that are all the same. Stuffed with keywords, empty of substance. On Claude they get ignored or merged with similar competitor sheets. Solution: every product/service deserves a page that says something true and specific. If you don’t have 1,000 sensible words to say about a product, you don’t know it well enough.
The blog with 400 articles of 600 words. Quantity without depth. Claude reads each one as an isolated fragment and none carries authority. Better 30 guides of 4,000 words than 400 posts of 600.
The “company profile” PDF behind a password or a form. If your best knowledge is locked away, Claude can’t read it, and neither can the sommelier who wants to upload it into their project. At least open up the technical sheets and the informational white papers.
The long guide without an opening summary. Length helps, but if Claude has to read 6,000 words to figure out what you’re talking about, the first section is wasted. Always open with the point, then develop it.
Here’s what you can do now
- Identify the 3-5 topics on which your company has real and differentiated knowledge. Not the topics “that are trending”. Yours.
- For each of those topics plan a guide of 5,000-10,000 words, written by someone who truly knows, with specific examples, internal data honestly disclosed, real anonymized cases.
- For every main product/service create a dense technical sheet of 1,500+ words, with a publicly downloadable PDF.
- Compare with the 3-5 competitors Claude cites in your industry: how much long content do they have? How deep is it? That’s your starting line.
- It’s not a magic factor. On ChatGPT or Perplexity, where the chunking mechanisms are different, the advantage of long content fades. But on Claude it weighs, and Claude is the tool of those who work on documents for a living — sommeliers, lawyers, doctors, consultants, researchers.
The basic logic is this: if your professional client uses Claude to work on their subject matter, and you are the supplier of that subject matter, you can’t show up with fragments. You have to show up with a body of knowledge.
In this series I’m telling how each AI platform has its own grammar. In the upcoming articles we’ll look in detail at the differences in Claude Projects as a working environment for professionals, and then at how ChatGPT and Perplexity handle long content in radically different ways. Visibility in AI answers isn’t played with a single piece of content across all platforms: it’s played by recognizing the specific rules of each one and equipping yourself accordingly.
