# When not to use AI for a page

A model is useful when the material exists and needs shaping. It is useless, and sometimes dangerous, when the material is something only you have: a customer's result, an opinion you hold, a decision you made, a promise you are accountable for. Six pages fall into that category. Here is the test, the six, and what a model can still do for them.

Updated 2026-09-15 · Source: https://porteur.ai/guides/when-not-to-use-ai-for-content

## The test, in one question

Ask this before you open a chat: if this page were wrong, who would be embarrassed, and about what?

- Nobody, or the reader learns something slightly off: a model can draft it from your facts.
- A customer, because you quoted them: write it yourself.
- You, because you claimed something you do not believe: write it yourself.
- Your company, because the claim is a promise about money, data or safety: write it yourself, and have someone read it.

The line is not about quality of prose. A model writes clean sentences. The line is about accountability: who stands behind the claim, and whether the claim came from anywhere.

## The six pages to write yourself

| Page | Why a model cannot write it | What breaks if it does |
| --- | --- | --- |
| The case study | The figures came from one customer and exist nowhere else | An invented or rounded result, published under a real company's name |
| The opinion piece | It has no opinion, and it will produce the consensus one | You publish a position you do not hold and cannot defend in the replies |
| Your method | You invented it. There is nothing in its training about how you work | A generic process with your name on it, which your customers can tell is not what you do |
| The security and trust page | It does not know what you encrypt, where data sits, or who can see it | A claim about data handling that is wrong, which is a legal problem and not an SEO one |
| The pricing page | Prices, limits and what is included change every quarter and only you know them | A plan that does not exist, a limit that is wrong, a refund promise you did not make |
| The founder's story | It was not there | The page every visitor reads to decide if you are real reads like everyone else's |

> Four of these six are the pages a buyer reads immediately before paying. They are the worst possible place to save twenty minutes.

## Why a model cannot write them, in one paragraph

A model produces text from patterns in what it has read. Facts it was never given are filled with plausible substitutes: figures, features, dates, policies. That is not a flaw to be prompted away; it is how the thing works.

The six pages above are made almost entirely of facts it was never given. Your customer's number. Your reasoning. Your infrastructure. Your prices. Your history. Hand it those pages and it does what it always does: it produces the average version, confidently.

It also has no first-hand experience and no opinion, which is what the opinion piece and the method page are made of. It can imitate having one, which is worse than not having one.

## What it can still do for those pages

The rule is not “keep the model away”. It is “keep the material yours”. Once the facts and the judgement exist, the model has work to do.

1. **Interview you** Ask it for the twenty questions a reader would ask about your method or your story. Answer them out loud, paste the answers back, and let it order them into sections.
2. **Structure the case study** Give it the customer's words, the before figure, the after figure and what was done. Ask for the standard shape: who, the problem in their words, what was done, the result, the quote.
3. **Draft from your notes** Your security page starts as a list of true statements from your engineer. The model turns the list into sentences. You check every one.
4. **Write the questions, not the answers** For a pricing FAQ, ask it what buyers ask. Answer each one yourself. The questions are pattern; the answers are policy.
5. **Edit, once you have written** Cut the length, plain up the register, catch the repetition. This is where it is genuinely good and where nothing is at risk.

## The grey zone, and how to decide

Most pages are not in the six. Guides, explainers, comparisons, documentation and how-tos are drafted with a model every day without a problem, as long as the facts came from somewhere.

- A guide to a public subject: fine, with a brief and sourced facts.
- A comparison of your product with a rival: draft the structure, write the claims yourself, and check every feature on the rival's own site the day you publish.
- Product documentation: fine, from the actual behaviour, and only if someone who has used the feature reads it.
- An integration page: fine for the shape, not for the setup steps, which must be the ones that actually work.
- A page family generated per city, per tool or per combination: this is where the grey zone ends. Google's spam policies name scaled content abuse, which is producing many pages primarily to rank rather than to help, whoever produces them.

## What it costs when you get this wrong

The SEO cost is the least of it. A thin guide ranks nowhere and you lose an afternoon. The other costs are larger and slower.

- A wrong claim on a pricing or security page is a promise you did not make, and the person reading it is about to pay you.
- A case study with a number the customer did not give you ends the relationship, and they will say why in public.
- An opinion piece you cannot defend is answered in the replies by someone who has done the work.
- A founder page that reads like everyone else's removes the one advantage a small product has over a large one.

Write those six yourself, badly if necessary. A plain page in your own words outperforms a polished page that is not true.

## Questions

### Is it against Google's rules to use AI for any of these pages?

No. Google rewards quality content however it is produced, and does not penalise AI assistance as such. The reason to write these six yourself is accuracy and accountability, not policy.

### Can I use AI for a case study if the customer approved the text?

Yes, once the figures and the quote are theirs. Give the model the real numbers, the real words and what was done, and let it arrange them. What it must never do is supply the result.

### What about product documentation?

Draft it from the actual behaviour, then have someone who has used the feature read it. Docs are the pages developers and assistants quote, so a step that does not work is expensive in support time.

### How do I write an opinion piece with a model, then?

Have it interview you. Ask for the ten questions a sceptic would put to your position, answer them out loud, paste the answers back, and ask for an outline. The opinion stays yours; the ordering is the model's.

### My competitor generates hundreds of pages. Should I?

Look at whether those pages hold anything of their own before copying the strategy. A family where every page carries real data, a worked example or a tool earns its place; a family where only a name changes is the pattern Google's scaled content policy describes.

## Read next

- [How to write a case study page that ranks and convinces](https://porteur.ai/guides/how-to-write-a-case-study-page): Use this fixed shape to write a case study page that ranks on buyer searches and helps them decide. With examples, schema and a checklist.
- [Trust signals on a product site: what a stranger checks before paying](https://porteur.ai/guides/trust-signals-on-a-product-site): What a stranger checks in the first minute, the pages that answer each doubt, how to show proof without noise, and a checklist you can ship today.
- [The SaaS pricing page: what it must say for the buyer and for search](https://porteur.ai/guides/saas-pricing-page): Your SaaS pricing page must sell to buyers and rank for pricing searches. Here is what to include, how to mark it up, and what to track.
- [AI-written content and Google: the policy, in Google’s words](https://porteur.ai/guides/ai-content-and-google): Google does not penalise AI writing as such. It acts on pages made at scale with nothing of their own. Here is the policy and where the line sits.
- [A writing workflow with AI that does not produce filler](https://porteur.ai/guides/ai-content-workflow): Five steps: the facts, the brief, the draft, the human pass, the read-back. The shapes to copy, and the failure each step prevents.
- [E-E-A-T](https://porteur.ai/glossary/e-e-a-t): E-E-A-T means Experience, Expertise, Authoritativeness and Trustworthiness. What it means for a small product site, and how to show it.
- [Scaled content abuse](https://porteur.ai/glossary/scaled-content-abuse): What scaled content abuse means in Google’s spam policy, how to avoid it on a small site, how to check risk, and what to fix fast.

Paste your URL and the free check reads your pages as a crawler and a buyer see them, in about thirty seconds, and names the ones that say nothing only you could say. Free check: https://porteur.ai/