A blog post written with AI works for Google when it answers a search intent, puts the answer up front, draws facts from the business and is written without clichés, in the company's own voice. This requires a briefing with form of address, region, services and taboo words, a draft with subheadings and FAQ, and a review before publishing. Google does not judge the production method, it judges usefulness.
How to use ChatGPT is covered in plenty of guides, including our post Writing blog posts with AI in seven steps. This page covers what matters between those steps: what quality a post needs to be taken seriously by Google and readers, what a briefing that produces this quality looks like, and which mistakes slip through most often in practice.
What Google expects from an AI-written article
In February 2023, Google stated publicly that content is judged on its quality, regardless of whether a human or a machine created it. The deciding factors are the criteria for helpful content: experience, expertise, reliability, and whether the page was written for people rather than for the search engine. Whether AI text gets penalised is covered in detail on the page Does Google penalise AI text?. For the workflow, this means: the risk is not where the text comes from, it's a text without substance.
Five quality criteria shared by Google and readers
- The answer comes first. Anyone searching for "What does a heat pump cost in a detached house" wants a range in the first paragraph, not an introduction about climate change.
- The text knows something only this business knows. A project example, a regional note, a price, an experience from everyday work. That's the part no model can supply.
- Figures are backed up or marked as an example. "Sample calculation" is fine, a made-up statistic is not.
- One search intent per article. A post that covers costs, process and funding all at once doesn't properly answer any of the three questions.
- The language fits the business. Formal or informal address, Swiss Standard German, no clichés, no marketing language the owner would never use on the phone.
The briefing: six fields are enough
A briefing is the instruction that comes before every draft. Six fields are enough for everyday use: target audience, the specific question, form of address, region, the business's services (and what it doesn't offer), and taboo words. The full template with a filled-in example is at Briefing for AI texts. Once you've written the briefing properly, you can reuse it for every further article and just change the question.
The process in four blocks
| Block | What happens | Who |
|---|---|---|
| Briefing | Define question, target audience, rules | Human, once, plus a question per article |
| Draft | Text with subheadings, answer up front, FAQ, internal links | AI |
| Enrichment | Add project example, price, regional note | Human, a few sentences |
| Review | Facts, form of address, spelling, clichés, links | Human |
The third block is the one most people skip. It turns a correct text into one that genuinely comes from your business. Two or three sentences are enough: "Last year, in a terraced house in Amriswil, we..." or "In Thurgau, such systems are subject to...". AI can't write these sentences, because it doesn't know the business.
Checklist before publishing
The short version for everyday use, the detailed twelve-point list can be found at Checking AI blog posts:
- Does the first paragraph answer the question from the title?
- Are all statements about services, prices and region correct?
- Is every figure backed up or marked as an example?
- Is the form of address consistent throughout, no ß, no dashes used as punctuation?
- Does the text contain at least one sentence that could only come from your business?
- Do the internal links lead to existing pages?
- Does the FAQ section contain genuine follow-up questions rather than repetitions?
For the fine-tuning of title, meta description and subheadings, there's a separate checklist at Optimising blog posts for Google.
Typical mistakes and how to spot them
| Mistake | How to spot it | Fix |
|---|---|---|
| Invented service | Text mentions an offer the business doesn't actually provide | Add "We don't offer" to the briefing |
| German spelling | Eszett, "Fahrrad" instead of "Velo", prices in euros | Set Swiss Standard German in the briefing |
| Form of address changes | First paragraph uses "Sie", FAQ suddenly switches to "Du" | State the form of address explicitly, check for it specifically during review |
| Introduction without an answer | First paragraph explains why the topic matters | Require the answer within the first three sentences |
| Stock phrases | "In today's world", lists grouped in threes | Put taboo words in the briefing, see the list under Spotting AI stock phrases |
| Three topics in one text | Subheadings jump between costs, process, funding | One question per article, the rest in separate posts |
| General figures | "Studies show", "up to 40 percent" without a source | Remove or rephrase as an example calculation |
A real-world example
A painting business wants an article on the question "How much does it cost to have a flat painted". Without a briefing, the AI delivers an introduction about the importance of fresh colours, German price-per-square-metre figures in euros, and advice to get several quotes. With a briefing ("Sie" form of address, Eastern Switzerland region, prices as an example calculation per room, taboo word "high quality"), the result is a draft that states a price range in the first paragraph, lists the factors in a table, and answers the question about covering work in the FAQ. The owner adds two sentences about a project in Arbon and approves it. The difference isn't the tool, it's the briefing and those two sentences.
Frequently asked questions
Do I need to heavily rewrite the AI text so it works for Google?
Don't rewrite it, enrich it. Google doesn't need different sentence structure, it needs content that couldn't exist without your business: prices, examples from projects, regional details. You add those parts, and leave the rest as it is if the briefing was good.
Is ChatGPT alone enough to write blog posts regularly?
For individual posts, yes. For a weekly rhythm over months, the effort per post becomes the problem: adding the briefing, checking the draft, formatting, copying it into the website, finding an image, sharing it. From around one post a week, a process that combines these steps pays off.
How do I tell that an AI article is serving the wrong search intent?
Google the question the article is meant to answer and look at the first five results. If those are guides and your article is a product page, or the other way round, the intent doesn't match. A second sign: the article half-answers three questions instead of fully answering one.
How long should an AI-written blog post be?
As long as the answer requires. For a specific customer question, 600 to 1'000 words is a good range, since that leaves room for the answer, an example, a table and an FAQ. Artificially lengthened texts stand out during review because of repetition.