Tuesday, 10:10 PM
An air conditioning business owner asks ChatGPT: "Write me text for the home page." He gets: "We offer professional, high-quality air conditioning solutions, with personal and reliable service for every customer." He reads it once and feels something is off, but can't say what. Then it hits him: this sentence could be pasted onto the website of any AC installer in the country, and nobody would notice.
The text is fine. It just isn't his.
That is the real problem with writing website content with AI. The tool writes fast, in decent language, with no mistakes. But without material from your business, it writes the most predictable thing possible, and what's predictable doesn't sell anything. This guide explains why that happens, what Google officially says about it, and how to get text out of the tool that sounds like you.
In short
- Writing website content with AI fails when you give the tool only a topic. Without facts, a voice and an opinion, it fills the gap with the internet's average phrasing.
- According to Google, AI content isn't disqualified because of its source. What may violate the policy is lots of pages with no value to users.
- The fix is a method: raw material from the business, a style guide, and an editing round. Only you can supply the material.
- At the end of the article there's a "generic text detector" checklist you can run on any draft, yours or a tool's.
Why does AI-written text sound generic?
A language model writes based on patterns. When it gets "write text for the home page of an air conditioning company," it knows nothing about your business, so it picks the most likely continuation. The result is the average. (If you want to understand how this works under the hood, we have an explainer on what an LLM is.)
Three things are missing, and they're missing because of the request, not the tool:
Facts
- Prices, arrival times, what's included and what isn't
- Questions customers really ask
- Without it: "quality and reliable service"
Voice
- How you talk to a customer on the phone
- Words you love and words you hate
- Without it: the tone of a faceless company
Opinion
- What you think customers get wrong
- Where you disagree with competitors
- Without it: "every option has pros and cons"
The word that sums it up is specificity. Generic text is text you could hand to any business in the industry without changing a word. Good text is text only you could write.
What does Google say about AI-written content?
This is the question every site owner asks, so let's answer it from what Google itself wrote, not from rumors. We checked the official Google Search Central documentation.
Myth"Google penalizes all AI-written content"
In practiceIn the page "Google Search's guidance about AI-generated content," Google doesn't say it's prohibited in itself. It warns against using AI tools to produce many pages without adding value for users, and classifies that under its spam policy on scaled content abuse.
Myth"If I don't disclose it's AI, that's deception"
In practiceGoogle recommends thinking in terms of Who, How, Why: who created the content, how it was created and why. It writes that sharing information about how content was produced can give readers context. This is a recommendation, not an obligation to label every paragraph.
Myth"If the tool wrote it, I can publish it as is"
In practiceGoogle notes that AI output may contain inaccuracies, known as hallucinations, and writes that it is essential to manually review and check AI-generated content before publishing. We have a separate article on this: What is a hallucination in AI.
So what does Google look for? In its page on helpful content it offers self-assessment questions, and several of them fit our topic exactly:
- Does the content provide original information, reporting, research or analysis?
- Does it provide substantial value compared to other pages in search results?
- Was it written or reviewed by someone who really knows the subject?
- Does it reflect first-hand experience?
Notice that none of these questions asks "who typed it." They all ask what is in the text. That's why, under the scaled content abuse policy, what bothers Google is the mass production of pages meant to manipulate rankings rather than help users, not the tool you used. A small business that publishes one article a week written from real material is on a completely different side of the line from a site that pumps out a thousand "plumbing services in [city name]" pages at the push of a button.
What puts a site at risk
Content at scale, with no value
- Dozens of pages that change only the city name
- Text generated and published without reading
- Unchecked information, including prices and facts
- The same structure on every page
What Google asks for
Content made for people
- One page for each real question
- Experience and details only you have
- Checked and edited by a person who knows the subject
- Clear who wrote it and why
The method: raw material, voice, editing
If the problem is that the tool doesn't know you, the solution is to introduce yourself. We work in three stages, and two of them are yours.
From a generic request to your text
- 1. Raw materialCustomer questions, prices, your photos, opinions
- 2. Style guideHow you talk, what's off limits
- 3. DraftThe tool writes and flags what it didn't understand
- 4. Editing and approvalFact check, generic detector, publish
Stage 1: raw material, five sources you already have
You don't need to be a writer. You need to pass material along, and that can be done with voice messages.
- Customer questions. What people ask by phone, on WhatsApp and by email. A page that answers a real question will beat a page that tries to "cover the topic."
- The logic behind the price. Not necessarily the price itself, but what affects it and why. It's the first thing a customer wants to understand, and few websites explain it.
- Your own photos and details. A photo from the field, a description of a job you did this week, a mistake you saw at a customer's place. That's the first-hand experience Google mentions.
- Opinions. "I don't recommend X, because..." An opinion is the one thing the tool can't know for you, and if you ask it to invent one, it will belong to nobody.
- Examples of writing you liked. A post of yours that worked, a message you wrote to a customer, an explanation you heard yourself give.
Stage 2: a one-page style guide
A short document you attach to every request. It answers seven questions:
- Who are we writing for, and what state are they in while reading?
- How do we sound: direct, warm, dry? Formal or casual, "we" or "I"?
- Which words do we like, and which are banned? (A blacklist of "solutions," "quality," "professional," "personal" without proof.)
- How short? Sentences, paragraphs, headings.
- What's always on the page: price, timing, how to get in touch.
- What must never be invented: numbers, quotes, testimonials, names.
- Three sample paragraphs, written or approved by you.
If you want a starting point for phrasing the requests themselves, we have 45 ready-made prompts and a method for writing a prompt. Here we're dealing with what sits beneath the prompt: the material.
Stage 3: editing, because a draft is only a draft
Our rule: the tool writes, a person decides. A good editing round includes a fact check (every number, price and name), running the generic test below, and replacing at least one sentence with a personal opinion. And if all that sounds like work, you can ask the agent to do the first round and flag only what looks suspicious.
Before and after: the same page, two requests
The example below is made up: an imaginary air conditioning business, with numbers created for illustration. The first request was "write me text for an AC installation page." For the second request we attached raw material and a style guide.
Before: a generic request
"We specialize in professional air conditioner installation. Our experienced team provides personal, high-quality service, with strict adherence to the highest standards. Contact us for a no-obligation quote and we will be happy to be at your service."
- Could go on any installer's website
- No price, no timing, no explanation
- "The highest standards": what does that mean?
After: raw material and style
"Most customers call and ask right away, 'How much for an AC?' The answer: it depends on the room. A 25 m² living room usually needs a bigger unit than a kid's bedroom, and the price also changes with the length of the piping and the wall we have to reach. We do a ten-minute walkthrough, measure, and only then quote. And if someone gives you a price over the phone without seeing your apartment, ask them what happens if the piping turns out longer than expected."
- Answers a real customer question
- Explains what affects the price
- Has an opinion, and the tone of a real person
An illustration. The business, the texts and the details are made up. The second version was written from raw material fed to the tool: a common question, pricing logic and an opinion.
What changed? Not the tool and not the model. What changed is what it received. In the second version every sentence comes from material the owner supplied: the common question, the pricing explanation, the recommendation. The difference isn't "nice style," it's that there's something to read.
The table that sums up the difference
| Generic text | Text based on raw material | |
|---|---|---|
| Starting point | A topic only | Customer questions, facts, opinions |
| What the reader learns | What they already know | Something they didn't know before |
| Voice | The tone of a "company" | Sounds like a person you can talk to |
| Risk of errors | Filler that sounds right | Specific factual mistakes, so you check |
| What you do | Copy and publish | Provide material, edit, approve |
| After publishing | One of many | A page that "belongs" to you |
Generic text detector: a checklist of 10 signs
Put a draft in front of you and tick every sign you recognize. The higher the count, the more generic the text. Each sign is phrased so you can check it in a single read.
How many generic signs does your draft have?
Rules of thumb: 0 to 2 signs is text you can publish after light editing. 3 to 5 means going back to the raw material stage and adding more. 6 and up is probably text written from a topic alone, and it's better to start over with material.
Still: when not to use AI for writing?
We don't think everything should be written with AI, and that's worth saying openly.
- "About" pages and your story. If the tool words your story, it will look like any business's story. Write it yourself, and use the tool only for polish.
- Health, finance and legal topics. Google stresses that on such subjects, writing or review by an expert matters especially. Here the tool can help with preparation, not replace the expert.
- Reviews and testimonials. Never generate them. It isn't only a matter of ethics, it's simply a lie to the customer.
- Pages at scale. If the idea is to produce dozens of similar pages at the push of a button, stop. That's exactly the situation in which Google warns about scaled content abuse.
The most common mistake we see isn't "using AI," but skipping the editing stage. Anyone who reads every draft once against the list above is already ahead of most websites.
And what about the agent?
Everything we've described can be done manually, in a conversation with ChatGPT. It works, but you have to paste the material and the style guide again every time. An AI agent connected to your business keeps them permanently, writes a draft, uploads it to the site as a draft (not as a published page), and waits for your approval. The publishing itself always stays with you. If you run a WordPress site, the page on managing a WordPress site with an AI agent details ongoing maintenance, and the map of AI tools for small businesses covers more tools and when they're enough.
Frequently asked questions
What is writing website content with AI?
Writing website content with AI means using an artificial intelligence tool, such as ChatGPT, to prepare drafts of pages, articles and product descriptions. To keep the result from being generic, you need to feed the tool material from the business: customer questions, facts, opinions and a style guide, and then edit and check the draft before publishing.
Does Google penalize AI-written content?
According to the official Google Search Central documentation, using AI is not prohibited in itself. What may violate the spam policy is creating many pages without adding value for users (scaled content abuse). Google also stresses that it's important to manually review AI-generated content before publishing.
Do you need to disclose that content was written with AI?
Google recommends explaining to readers who created the content, how and why (Who, How, Why), and writes that sharing information about how content was produced can give context. That recommendation doesn't require labeling every paragraph. If a page always names the person responsible for it, and AI was used substantially, you can mention it briefly.
How do you tell that a text sounds generic?
Check whether you can swap the business name for a competitor's without breaking any sentence, whether it has numbers or details you can verify, whether it has an opinion, and whether it answers a question a customer really asked. The checklist in the article gathers ten such signs.
How much raw material do you need to get started?
Less than it seems. Ten real customer questions, a short explanation of your pricing, three opinions and two paragraphs you wrote yourself are enough for one good page. You can record a voice message and let the tool transcribe and organize it, then edit.
Can you let an AI agent write and publish on its own?
You can let it write and prepare a draft on the site, but we recommend that publishing stay subject to your approval. That way you check the facts and decide what goes out in the business's name.
