Generic AI writing passes every check except being about anything.

Generic prose is grammatical, accurate and well organised. Its failure is that it is indistinguishable from the other twelve pages answering the same query, which in a search result is all the difference that matters.

It is also what a model produces by default. Asked about quality assurance, it draws on every text about quality assurance and lands on the common denominator, which is technically correct and adds nothing to what the reader already knew.

The common denominator is not worthless as a starting point. It is worthless as a finished page, because the reader's question was specific and the answer is general.

So the job is not to make generic prose correct. It is to make it yours, and that requires material the model did not have: your numbers, your constraints, your failures.

Generic writing is full of nouns that belong to nobody. The team, the business, users, stakeholders, modern organisations. Nothing in the sentence says which team, which users, or what they were trying to do.

The second marker is the verbless claim: efficiency matters, security is important, consistency drives trust. These are true and unfalsifiable, and a page built from them can be read without the reader learning a single fact.

Take a paragraph and replace your product name with a competitor's. If it stays true, it is a category paragraph. If it becomes false or absurd, it is doing work for you.

Run it on the whole page, then on the opening and the close. Most pages survive the swap for about 80% of their text, which is a fair estimate of how generic they are.

The test is brutal and it is also cheap, which makes it one of the few reviews worth running on your own copy before someone else does. Run it on two competitors' pages too, and see which paragraphs are interchangeable in both directions.

Anything that survives in both directions is the floor of your category. You can leave one paragraph of it for orientation, but it cannot be the argument.

A number you measured, a screenshot, a client constraint, a failure mode you have personally hit. None of it is expensive to add, and none of it can be generated without being supplied.

Specifics also do the persuasive work that adjectives are asked to do and fail at. Saying we found eleven of the nineteen samples failed on a single check is more convincing than saying our rigorous process is highly effective.

Start with the three questions a reader in your category always has, and answer each with something only you could answer. That is often enough to differentiate a page completely, without touching the structure.

If you cannot answer any of them with your own material, the topic may not be one you have standing to write about yet. That is a useful finding rather than a failure.

It enters in the brief. A brief that names a keyword and a word count but no reader, no claim and no evidence produces a page that can only be assembled from what is already public.

It enters in the outline. Headings like Why it matters, Key benefits and Best practices are a template, and a template filled by a model reproduces the template's emptiness exactly.

It enters in the voice guidance. Instructions to be professional and authoritative push the draft toward the register that reads like everything else, because that register is the average.

Fix the earlier steps and the writing mostly fixes itself. Briefs that specify a reader, a claim and one piece of original evidence produce sharply different drafts from the same model.

The opening sentence carries more weight than the rest of the page combined, because it decides whether there is a second. The generic opener spends it on context the reader already has.

Compare a category opener with a specific one. Quality assurance is essential for modern software teams. Against: we ran the same checker over four thousand generated files and 6% failed on a key that was never used.

The specific opener also sets an expectation the page has to meet, which is a feature. It commits you to evidence and makes padding obvious to the reader and to you.

Rewrite the opening last, once you know what the page concluded. It is much easier to promise what you have already delivered than to invent a hook and then go looking for support.

Learn more, get started, book a demo. These are the calls to action that survive the swap test perfectly, which is why they convert badly: they ask the reader to want something they have not been shown.

A specific action names the thing, the effort and the outcome: upload a file and see the findings in ten seconds. That sentence tells the reader what will happen, and it can only be written by someone who knows what does.

If the honest action is small, say so. The cheapest conversions come from the smallest first step, and describing it precisely makes it easier to take.

Reserve the big ask for readers who have already received something. A request to talk to sales works far better after a page has given the reader a result than before.

Take one generic paragraph. Underline every noun that could belong to another company and every adjective that could apply to any product. That is the material to replace.

For each abstract noun, ask what actually happened. Not improved efficiency but cut review time from two days to four hours. Not better collaboration but the second reviewer saw the diff before it merged.

Keep the sentence structure while you do it. The goal at this stage is substituting content, not rewriting rhythm, and changes made one at a time are easier to keep.

Then read the paragraph against the swap test again. A paragraph that fails for a competitor has earned its place; one that still passes needs another fact in it.

Some claims cannot be supported and should be removed rather than dressed up. Market-leading, industry-standard and best-in-class cannot be verified even by the company making them, and a reader who works in the category knows it.

Some sections exist for navigation rather than persuasion: definitions, glossary entries, comparison tables. There, generality is appropriate because the reader is checking a fact rather than being convinced.

Mark the difference. A page that mixes navigation and argument without signalling which is which reads as generic even where its argument is strong.

Cut rather than soften. A removed claim costs nothing; a vague version of it occupies space a specific one could fill, and it teaches the reader to skim the rest.

Add two lines to every brief: the reader, in one sentence, with the situation they are in; and the one thing this page will state that no other page does. If the second line cannot be filled, the page is not ready to be written.

Then run the swap test before anyone else reads the draft. It takes a minute and finds the paragraphs that read like a brochure, and a brochure is what generic writing is.

Keep a file of specifics: measurements, screenshots, support tickets, odd edge cases. It becomes the raw material that makes the next page impossible to swap, and it outlasts any particular model.

The comparison tool on this page does the mechanical part: it shows whether a rewrite actually lowered the stock phrasing or only rearranged it. The judgement about what is specific stays with you.