Technote

Working with AI Advanced

Competitor analysis with AI — open the page and verify

Summarising structure across several articles and spotting coverage gaps is genuinely useful work for AI. Every factual claim about a competitor, though, must be checked by opening the page — it will describe pages it never read, in detail.

Competitor content analysis eats time. Opening the top ten results one by one, working out how each is structured, noting what they cover and what they skip — half a day goes easily. AI shortens that considerably, provided you give it the actual text. That condition is very nearly the whole article.

Where it is strong: structure and coverage gaps

Paste in the body text of several articles and ask for “a table of the topics each piece covers, with a count of how many pieces cover each topic”. The results are good. A person doing this loses consistency around the fifth article; a machine counts the twentieth by the same rule as the first.

The gaps in that table are the point of the exercise. A topic covered by nine of ten pieces is table stakes and belongs in yours. A topic covered by none of them is an opportunity — though whether it really is one needs a separate judgement, which comes later.

Asking about tone and format patterns is useful too. “How do the introductions in these pieces typically open?” gives you material for deciding where to deliberately diverge.

What it invents: every fact about the competitor

Give it a URL and ask it to “analyse this page” and it will produce an analysis whether or not it can read the page. Heading structure, a list of subheadings, an approximate length, even a quotation supposedly from the page — all delivered fluently. Those are what an article on that subject would plausibly contain, not what that page contains.

The danger is that the output is plausible. A wildly wrong answer gets caught; a sentence like “Competitor A publishes its pricing” goes straight into the plan unchecked. And then the plan becomes the basis for your own pricing decision.

With the source text it is summarising; without it, it is composing

The procedure: paste, verify, record

Four stages — skip the third and the first two become guesswork

Collecting source text means actually opening each page and copying the body. It is tempting to automate, but seeing each page once here makes the verification step far quicker later.

Verification means reading the summary line by line and asking whether the basis for that sentence exists in the text you pasted. Anything unsupported gets deleted. Above all, copy quotations from the source yourself — a quotation the summary produced is most dangerous when it is most convincing.

The gap list is what survives verification. It is shorter than what you started with, but every line on it has been checked, so it can go straight into planning.

Finding a gap is not a reason to fill it

A topic nobody covers can mean two things: everyone genuinely missed it, or there is no reason to cover it. When not one of ten pieces mentions something, suspect the second explanation first.

And even where the gap is real, whether you have grounds to fill it is a separate question. On a subject you have never measured and never had to deal with, spotting the gap does not put you in a position to write about it better than anyone else. The conclusion of competitor analysis should be not “let’s write this” but “do we have grounds to write this”.

Placing those gaps into your own content structure is covered in the designing a content hub series, and the wider limits of AI in this kind of work live in the AI archive. If the site that has to carry the plan needs work first, see our optimization program.

More on this topic

All technotes
₩270,000 · Join the program