It's an unpleasant discovery: you ask ChatGPT a question you're the answer to, and the answer names your competitor. Not because they're better at the trade, but because they're more readable for the machine. The good news is that the cause almost always sits in a short list, and that you can check that list yourself. How to run that first test is covered in how to check whether your business appears in ChatGPT.

01 / structureThe competitor is easier to cite

AI systems don't cite pages, they cite passages. A page that has the question literally in a heading, the answer directly below it, and makes that answer readable on its own, produces usable passages. A page where the same answer is spread across a narrative produces nothing to cite.

The check: open the page of your competitor that's named as a source, and put it next to your own page on the same subject. Read both as if you only get to see the first two paragraphs, because that's roughly what an AI system cuts out. Look at where the answer sits. If it's in the first paragraph for the competitor and somewhere halfway through for you, you've probably already found the cause.

That difference in readability isn't a matter of writing talent but of structure, and structure can be fixed without changing the content.

02 / substantiationThe competitor gives figures with sources

Texts with concrete figures, source references and quotes from experts are cited more often than texts with only claims. An AI system building an answer would rather lean on a page that states where the knowledge comes from.

The check: count on your competitor's page how often a figure, a year or a source appears. Do the same for yourself. "We have a lot of experience" is nothing to a machine; a concrete figure with a provenance is something.

03 / third partiesThe competitor exists beyond his own site

This is a cause that's easily underestimated. According to analyses of AI citation patterns, brands are cited 6.5 times more often via third-party sources than via their own domain. Reviews, trade media, comparison sites and forums carry heavy weight in AI answers. A competitor who appears in three comparison articles and a review site turns up in answers without his own site needing to be cited.

You're not just competing with your competitor's site, but with everything the web writes about him.

The check: look in the source references of the AI answer to see whether they point to your competitor's own site, or to third parties that mention him. In the second case, the answer isn't "build a better site" but being present in the places where your market gets described. More on this in why AI refers to third parties more often than to your site.

04 / recencyThe competitor is more recent

AI systems factor in freshness. A page with a visible, recent date regularly beats an undated page with the same content, and a text from 2019 on a subject that moves is treated with suspicion.

The check: see whether your core pages state when they were last updated, whether that's recent, and compare it with the sources the answer does draw on. Outdated content isn't a cosmetic problem: for an AI system it's a reason to pick a different source.

Recency, incidentally, is the cheapest of the five causes to fix. Going through an existing piece, updating the outdated passages and making the update date visible costs a fraction of new writing work, and counts for an AI system as a fresh source.

05 / accessYour site keeps the crawlers out

The hardest cause: if your robots.txt blocks the crawlers of AI systems, what may not be read cannot be cited. This happens regularly, sometimes due to a default setting in a security package or a deliberate choice by a previous administrator.

The check: open yourdomain.nl/robots.txt and look for rules with GPTBot, PerplexityBot, ClaudeBot or Google-Extended after "Disallow". What you find there and the trade-off behind it is set out in allowing or blocking AI crawlers.

The five causes in one overview, with the signal and the first step for each:

Start at the bottom of the table: access is checked in two minutes and structure within a day, and together those two explain most cases.

One caveat to finish: sometimes there's no shortcoming and the competitor simply started earlier. In that case the list above isn't a diagnosis of what's wrong, but the worklist for what needs to happen. How we approach that work is set out at datadriftdigital.nl.

This text was produced with AI assistance and checked and approved by a human before publication.