Why does ChatGPT ignore our website when we have hundreds of pages?

You have paid for content for years. There are three hundred pages on the site, a blog fed every fortnight, and service pages for every variation of what you do. Then someone asks ChatGPT or Perplexity who to talk to about your specialism, and the answer names two competitors with a fraction of your output.

Rule out the technical causes first. If the assistants can read you and still do not name you, the cause in our experience is usually that most of those pages are thin. Not short, necessarily, but empty: they define the subject, say it is important, list some features and stop.

The research points the same way. In a Princeton and IIT Delhi study presented at KDD 2024, adding statistics to a page raised its visibility in AI-generated answers by about 34% and adding quotations from relevant people by about 41%, while keyword stuffing lowered it by about 8%. It was tested on a research engine, so treat it as directional. Google's own guide to its AI features, last updated in July 2026, makes the same distinction: content with "unique expert or experienced takes" counts for more than commodity content that "could originate from anyone".

#Is thin content the reason, or is it something else?

Check four other causes first, because each makes content quality irrelevant until it is fixed. An assistant cannot use a page it cannot reach or read, or a firm nobody else mentions.

Our guide to how AI crawlers read your website covers the first two checks in detail, and whether backlinks still matter for AI search covers the fourth.

#What do AI assistants count as thin content?

Any page that offers little a reader could not have guessed. Years of writing to satisfy search signals such as keywords and length left many B2B sites with long pages that say nothing, and an assistant needs a passage worth quoting.

#Definitions with no depth

Thin: "Cash flow forecasting is the process of estimating the money coming into and going out of a business over a period. It helps businesses plan."

Substantive: "Cash flow forecasting estimates the money moving in and out of the business week by week, usually over thirteen weeks. Unlike a budget, which sets targets for a year, a forecast is a working document you update every Monday. The point is to see the shortfall in week nine while you still have eight weeks to act on it: chase debtors, delay a purchase, arrange an overdraft. For most owner-managed firms the thirteen-week horizon is right because it covers a quarter's VAT and a full debtor cycle."

The second explains how it works, how it differs from a budget, why it matters and when it applies: a definition, a distinction, a use and a reason, all quotable.

#Five other kinds of thin page

• Lists with no analysis: "Five ways to improve cash collection: invoice promptly, chase early, check credit." True and useless until each point says why firms fail at it.
• Padding: "Cash flow is important for B2B firms. Many B2B firms rely on cash flow." The same idea twice.
• Keyword repetition: "Our cash flow software provides cash flow solutions for cash flow management."
• Boilerplate: "Cash flow is a complex subject. Working with an adviser can help." Swap the subject for any other and the page reads the same. That is the test.
• Near-duplicates: "cash flow forecasting", "guide to cash flow forecasting" and "best cash flow forecasting approach", the same ground with the words rearranged. The most common form of thinness on B2B sites, and the least noticed.

Feature lists, comparison tables with no verdict and case studies with no outcomes fail the same test; why ChatGPT never quotes our blog covers those.

#How does an assistant decide a page is thin?

Nobody outside the AI companies can see the exact process, but the pages we see quoted share five qualities, the same ones a sceptical buyer looks for. Density: "spending varies and depends on your needs" says nothing; a range, what drives it and a named source says something. No repetition: each sentence adds a new idea. Specificity: "you should use LinkedIn" is generic; "for firms selling to IT directors, LinkedIn advertising is worth testing, with a landing page built for that one audience" is specific. Support: "the return is typically eight to one", on its own, is a red flag. Coherence: the page sets out an argument and works through it. How assistants judge whether the writer knows the subject is covered in what E-E-A-T is and why AI cares.

#Why can a page rank on Google and still be left out of AI answers?

Because ranking and being quoted are related but different tests. Ranking helps: Google's AI features, Copilot and ChatGPT search all start from a search index. But Ahrefs found in March 2026 that only 38% of AI Overview citations came from pages ranking in the top ten, down from 76% in July 2025, because Google now runs several related searches for one question (query fan-out).

What we see in the answers we record is exclusion more than demotion: a thin page that ranks well is simply not quoted, while a more specific page further down is. We cannot see inside the assistants, so treat that as observation. For example, a broker ranking second for "commercial property insurance broker" with generic pages may be named less often than a competitor ranked fifth with detailed pages on underinsurance and insuring listed buildings. Why ranking first on Google does not guarantee a mention covers the wider problem.

#Why do duplicate pages do more damage than you think?

Because they blur which page is yours, and at scale they can break Google's rules. Google's scaled content abuse policy, introduced in March 2024, says: "Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users." It applies "no matter how it's created". Google's AI features guide adds that creating separate content for every variation of how people search, "primarily to manipulate rankings or generative AI responses", violates that policy, and its spam policies list "pages targeted at specific regions or cities that funnel users to one page" as doorway abuse. Mass location and segment pages with the same body are the obvious risk.

Duplication creeps in for sensible-sounding reasons: one page per segment, per town, per channel or per keyword variation, each with the same body. An assistant reading them cannot tell which page is the real one. When you update one and not the others, the site contradicts itself, which is worse than being thin; why ChatGPT describes firms wrongly shows the result.

Near-duplicates need not be word for word: rearranging a sentence does not make it new. Google's spam systems can act against a whole site; whether AI assistants mark down a whole domain in the same way is unproven. The practical cost is plainer: four pages saying one thing give an assistant one idea to quote.

#How do I find out which of my pages ChatGPT has cited?

Look in four places before you delete anything. Pages that assistants send people to, or fetch to answer questions, are the last ones to cut.

Our guide to finding out whether ChatGPT recommends your firm sets up the prompt checks.

#How to audit your content for thinness

#Step 1: Score your twenty most important pages

Choose them by traffic, AI use and commercial importance, and score each with the method in our guide to auditing content for AI assistants, so every page is judged the same way. For a quick first pass, ask: would I learn something, is it specific, and does every paragraph add something?

#Step 2: Find the duplicates

Search Console's page indexing report shows pages Google treats as duplicates, and tools such as Screaming Frog or Copyscape find near-duplicates. Look especially for the blog post, white paper and case study on one subject, and "beginner's" and "expert" guides that cover the same ground.

#Step 3: Compare with the pages assistants cite

Ask ChatGPT and Perplexity the questions your buyers ask, read the pages they cite, and compare their depth with yours on the same subject. Where you are thinner, you have found the gap.

#Step 4: Mark every page merge, deepen, keep or delete

Record the decision next to the score and the AI use from the previous section. That list is your work plan.

#How to fix thin content without rewriting everything

Merge the duplicates, deepen the pages that matter, delete the rest, in that order.

Merge duplicates. Combine three segment pages into one with a section per segment, or make them different in substance. "Software for small firms: five options under a set budget" and "software for enterprise: multi-region requirements" have different decision criteria, which makes them segmentation.

Deepen the pages that matter. Give each point on a thin "five ways to improve" page a proper paragraph or two: why firms neglect it, a real or labelled example, how long it takes, what changes when it is done. Where one thin page covers a big subject, replace it with a main guide and three or four deeper pages under it, as set out in organising content so AI assistants see you as the expert.

Delete what cannot be saved. Pages that do not rank, do not convert, are not used by assistants and duplicate what competitors do better should go, with the address redirected to the best remaining page. Keep old pages only if they are still accurate.

#What does a rebuild look like in practice?

A large site is not the problem when each page has a distinct job. arum, which advises banks, governments and major brands on collections technology, had "more than 500 projects in 23 countries, from offices in London, Atlanta and Sydney" to describe. The rebuild: "310 pages rebuilt across three regions, in weeks." "215 articles, research pieces and news items moved across, going back to 2014."

As arum's CEO, Jamie Waller, put it: "For most people, our website is the first time they meet arum. It now does us justice."

#How do we stop thin content coming back?

Apply a short checklist before anything goes live: the page answers a specific question a buyer asks, gives at least three usable points, includes a concrete example or sourced figure, differs clearly from what competitors have published, and adds something in every section. Length is not on the list. Concise content is specific in few words; thin content is generic at any length, as whether ChatGPT prefers long articles or short ones explains. Review the library once a year, merging what has drifted together.

#What to do this month

1. Run the four technical checks: robots.txt and CDN bot settings, page source, Bing, and mentions on other sites.
2. Pull AI referrals by landing page from GA4 and ChatGPT-User hits from your server logs.
3. Score your twenty most important pages with the audit method and mark each merge, deepen, keep or delete.
4. Run a duplicate check and merge or differentiate each group, starting with the subjects that matter most commercially.
5. Add real depth to the five thin pages with the most traffic: examples, reasons and figures with sources.
6. Give the pre-publication checklist to everyone who writes for you.

#Frequently asked questions

#Does having too many pages hurt my website in Google and ChatGPT?

Page count is not the problem; empty and duplicate pages are. Sites that break Google's spam policies "may rank lower in results or not appear in results at all". AI assistants have no published rule on page count, but a site where most pages repeat each other gives them little to quote.

#Should I delete old blog posts that get no traffic?

Check first whether assistants use them, in GA4 and your server logs. A post with no traffic, no AI visits, no links and a better page on the same subject should be merged or deleted, with the address redirected. A post that is accurate and says something no other page does can stay.

#Are location pages for every town bad for SEO and AI search?

Pages that swap a town name into the same text are what Google's doorway abuse policy describes. A location page earns its place when it says something true only of that place, such as the team based there. For most B2B firms, one page per real office is enough.

#Can AI-generated pages get my site penalised by Google?

They can if they are produced at scale and add nothing: Google's scaled content abuse policy applies "no matter how it's created". Use AI to draft a structure or tidy prose, then add what only you have: your examples, figures and judgement.

#Will merging duplicate pages hurt our Google traffic?

Sometimes, briefly, because three addresses become one. In our experience the stronger single page usually recovers within a few months. Redirect every old address to the merged page.

#How long before fixing thin content changes what AI assistants say about us?

Assistants that search the web live (ChatGPT search, Perplexity, Copilot, Google's AI features) can reflect a rebuilt page within weeks; answers from a model's memory change only when it is retrained, which takes months. ID-Pal is a dated example. Its case study reports: "UK buyer questions where AI names ID-Pal: none in May, two in August, four in September 2026." And: "Where a page has been rebuilt, ID-Pal appears. Where it has not yet, it does not, which tells us exactly what to build next." Read the ID-Pal case study.

Most firms have far more expertise than their website shows, and the fix is bringing it to the surface. To see how your firm is described today, book a free Growth Review: 30 minutes with David Adams and three fixes in writing. If the site itself needs rebuilding, that is Website That Sells.

#Sources