How to make AI content actually rank

The question everyone asks first is the wrong one. "Will Google penalise AI content?" has a boring answer — no. Google's guidance since 2023 has said the same thing in every rewrite: how content is produced is not the ranking signal. Whether it helps the person who searched is. A hand-written page that restates the first result adds nothing and ranks like it. A generated page that answers a question nobody else answered well ranks like that instead.

That does not mean AI content ranks by default. Most of it does not, and the reasons are consistent enough to list. Below is what separates the two piles, in the order the work actually happens.

1. Start from the query, not from the topic

The single most common failure is writing about a subject instead of answering a search. "Email marketing" is a subject. "How often should I email my list without people unsubscribing" is a search. The first produces a page that overlaps a thousand others; the second produces a page that can be the best result for one specific thing.

Before a word is written, you need three facts about the query: what the searcher wants to do next, what the current top results already give them, and what those results leave out. That last one is the whole opportunity. If you cannot name the gap, there is no reason for the page to exist and no amount of writing quality will manufacture one. Finding those gaps is its own method when your domain is the small one in the results.

2. Match the intent shape, not just the keyword

Every query has a shape the results already agree on. Some return step-by-step instructions, some return comparison tables, some return a list of options, some return a single definitional paragraph followed by detail. Publishing a 2,000-word essay against a query whose results are all short how-tos is not being more thorough — it is answering a different question.

• Informational ("how to…", "what is…") — direct answer in the first 100 words, then the detail underneath.
• Commercial ("best…", "… vs …", "alternatives to…") — a comparison structure with criteria stated up front, and a real recommendation at the end.
• Transactional ("buy…", "pricing…") — the product page, not an article. Do not write an article for these.
• Local ("… near me") — a location page with an address, hours, and service area.

Mismatched intent is invisible in a content brief and fatal in the index. It is also the failure mode AI writing amplifies most, because a model asked for "an article about X" will happily produce an essay for a query that wanted a checklist.

3. Give it something only you have

A language model knows what is already written down. If a page contains nothing else, it is by construction a remix of the corpus — which is precisely the "no added value" category Google's spam policies name. Anything that is yours breaks that:

• Your own numbers — conversion rates, delivery times, support ticket themes, price points, what a job actually cost.
• A worked example with real inputs and real outputs, not a hypothetical.
• A position you are willing to defend, including what you would not recommend and why.
• Screenshots, exports, or the actual sequence of steps from your own tooling.
• What went wrong. Failure cases are the single most under-supplied thing in commercial content.

This is also the practical meaning of E-E-A-T. Experience is not a badge you add in an author box; it is whether the page contains information that could only come from having done the thing.

4. Structure so the answer is extractable

Search results increasingly quote pages rather than link to them — featured snippets, AI overviews, assistant answers. Every one of those mechanisms needs to lift a self-contained span of text out of your page. Structure is what makes that possible.

1. One h1 that states the subject the way a person would say it.
2. h2 per real section; h3 for sub-points. Never skip a level for styling reasons.
3. A question-shaped heading gets a direct answer in the paragraph immediately under it — first sentence, no run-up.
4. Lists for anything enumerable, tables for anything comparable. A paragraph listing seven things is unquotable.
5. A short summary near the top for anyone who will not scroll, and for the snippet extractor.

Bolded pseudo-headings are the classic AI-output tell and they cost you. A bold paragraph looks like a heading to a reader and like body text to everything else, so the question it contains is never addressable.

5. Edit like an editor, not like a proofreader

Generated drafts fail in predictable ways, and they are not spelling. The edit pass that matters removes rather than polishes:

• Cut every sentence that restates the previous one in different words. Models pad; padding is the main thing readers bounce off.
• Cut the throat-clearing introduction. Most drafts do not start until paragraph three — delete the first two.
• Verify every specific claim, number and citation. Confident wrong facts are the real risk, not clumsy phrasing.
• Replace generic advice ("post consistently") with a decision the reader can execute ("twice a week; the third post per week did not change our traffic").
• Read the h2s alone. If they do not tell the story on their own, the structure is wrong.

6. Publish into a structure, not into a void

An article with no internal links is an orphan. Crawlers reach pages by following links, and the pages you actually want to rank commercially — your product and service pages — are reached mostly through the editorial pages around them. Every post should link out to the two or three pages it naturally supports, with anchor text that describes the destination rather than saying "here".

The same applies in reverse: a new post that nothing links to will be crawled late and forgotten early. On a store, the pages that most need those links are the category and product pages, which almost never earn external links of their own. Link to it from the hub page and from at least one older post the day it goes live.

7. Measure the right thing, and wait the right amount of time

New content on a young domain typically takes eight to sixteen weeks to settle into a stable position, and the first movement you see is usually impressions, not clicks. Judging a post after ten days tells you nothing except that it has been indexed.

• Week 1–2: confirm indexation and that the right page is the one indexed. Nothing else is meaningful yet.
• Week 3–8: watch impressions and the average position of the target query. Rising position with no clicks means the title and description need work, not the content.
• Week 8–16: judge the page. If it has settled outside the top 20 for its target query, the gap you identified in step 1 probably was not real.

The failure pattern to avoid is churn: rewriting everything every three weeks so nothing is ever old enough to have a result. Pick the target, publish, link it, and leave it alone long enough to learn something.

What it comes down to

AI writing changes the cost of producing a draft. It does not change what makes a page worth ranking. The work that decides the outcome is upstream of the draft (which query, what gap, what shape) and downstream of it (what did you cut, what did you verify, what does it link to). Teams that keep doing that work publish faster than before and rank. Teams that treat generation as the whole job publish more pages that nobody reads.

Frequently asked questions