Your content now has a second reader. Answer engines, from ChatGPT to Google’s AI Overviews, scan it, extract a passage, summarise it and cite it, or prefer another source. This reader does not read the way a person does.
In 2024, five researchers (Dentella, Günther, Murphy, Marcus and Leivada) asked seven of the most advanced language models comprehension questions about short, everyday sentences. The same questions, put to 400 human readers, served as the baseline. The results, published in Scientific Reports, showed that the models answered at chance level, changed their answer when the same question was asked again, and made errors of a different kind from those humans make. We draw a working rule from this. When a reader grasps the meaning of sentences poorly, structure becomes its main point of reference: the headings, where the information sits, how the text is divided into blocks. Poorly built content risks being miscited, or not cited at all. This exposes the limits of a common belief, that a few tags and an FAQ section are enough to “optimise for AI”. Tags help a machine find a block; they do not guarantee that the block gets it right.
What is an AI editorial audit?
An AI editorial audit extends the editorial audit. It covers the same body of content and asks one more question: can an AI extract each piece, summarise it and cite it without distorting it? It is an analysis of editorial performance from the point of view of that second reader.
It differs from two related exercises. An SEO audit checks that search engines can crawl and index your pages; it says nothing about what an AI will take from them. The AI footprint scan measures, from the outside, whether assistants cite your brand when your clients ask them questions. The AI editorial audit works from the inside: it searches your content itself for what makes it citable or not.
Three tiers of legibility
We read every piece at three levels, because an AI can fail at any of them.
Micro-level readability: the sentence
Immediate clarity, flow, no friction. A technical term left undefined, a double negative, a pronoun whose referent is unclear: a human reader recovers the meaning from context; a model may lose it. This is the only level that conventional readability scores measure, because they are based on sentence and word length.
Meso-level legibility: the block
Logical progression, one idea per block, transitions that hold. This is the level at which answer engines work: they break content into blocks and extract one to build their answer. A block that mixes two ideas produces a muddled answer; a block that does not stand on its own produces a wrong citation. The editorial usability rules that AI rewards include putting the key information first (the inverted pyramid) and dividing content into blocks.
Macro-level legibility: the system
Consistency with the brand’s voice, with the reader’s path and with the rest of your content. When two pages on the same site give two different definitions of the same service, or two different timeframes for the same engagement, an AI that reads both cannot tell which to believe, and its answer reflects the contradiction. Such contradictions are a form of editorial debt: nobody decided on them, and each one is paid for with every reading.
Citation value: a criterion to add
A content inventory usually classifies each piece by traffic, rankings and links. The AI editorial audit adds a column: citation value. A page can bring in no clicks and still be an asset, because an AI draws on it when answering.
We assess it using the three questions in our content inventory method:
- does the piece clearly answer a real question?
- can its expertise be attributed to an author and to sources?
- does its structure let a machine extract it without misrepresenting it?
The method in four steps
1. Choose the priority corpus
Your offer pages, in-depth articles and FAQs: the content an AI is most likely to draw on when a client asks it about your field.
2. Read at three levels
We read each piece sentence by sentence, then block by block, before setting it against the rest of the corpus. This is where we identify contradictions between pages.
3. Test under real conditions
We ask the leading assistants the questions your clients put to them, then compare their answers with what your pages say: passage reused, passage distorted, competing source preferred.
4. Classify and decide
Each piece receives a decision: strengthen, update or retire. The rules that emerge (page templates, terms to define, FAQ structure) go into your Brand Voice Framework, so that future content is legible from the start.
What the audit produces
- A content map, by type, status, citation value and legibility at all three levels.
- The list of contradictions between pages, each with the page that should take precedence.
- A prioritised list of the pieces to strengthen, update or retire.
- Writing rules for future content, built into the Brand Voice Framework.
Where it fits in our approach
At WeAreTheWords, this review is part of the audit that opens the Clarity and Growth engagements. The AI footprint scan, which measures your brand’s place in assistants’ answers, belongs to the Elite engagement.
Frequently asked questions about the AI editorial audit
What does an AI editorial audit check?
An AI editorial audit is an editorial audit that asks one more question of a body of content: can an AI extract, summarise and cite each piece without distorting it? It reads the content at three levels (the sentence, the block, the system) and assesses its citation value.
How does it differ from the AI footprint scan?
The AI footprint scan measures from the outside whether assistants cite your brand, with or without a link, or cite your competitors instead. The AI editorial audit works from the inside: it searches your content for what makes it citable or not, and identifies what needs fixing.
Why do AIs misread some content?
Because language models grasp the meaning of sentences poorly: in a study published in Scientific Reports in 2024, seven models tested on simple comprehension questions answered at chance level. Content whose key information is buried, whose blocks mix several ideas, or which contradicts another page on the same site gives them every opportunity to go wrong.
How does an AI editorial audit differ from an SEO audit?
An SEO audit checks that search engines can crawl and index your pages. An AI editorial audit examines what an AI will take from those pages once it has read them: what it will extract, how it will summarise it, whether it will cite it.
Making your content citable
Does your content answer the questions your clients put to AI, in a form AI can reuse? The Clarity engagement starts with that review.
To talk it through, book a call.