Ask an AI tool to write your main service page and you’ll have a tidy, confident draft in seconds. Ask it again for your closest competitor and you’ll get much the same page. That is the real problem when you write website copy with AI: the writing isn’t bad, it just could belong to anyone.
Buyers read your website to decide whether you understand their problem and can be trusted with it. Fluent sentences don’t answer that. Your prices, process, projects and opinions do, and no model knows them unless you tell it.
How to write website copy with AI: the short answer
Give AI the jobs that are about form, and keep the jobs that are about substance.
- AI does well: structuring your notes, suggesting alternative headlines and phrasings, tightening long sections and keeping terms consistent across pages.
- Only you can supply: the questions customers really ask, your prices and process, proof, opinions and first-hand detail.
- The workflow: gather source material, brief each page, outline around buyer questions, draft from your material, verify every claim, then edit for voice and test the page with a colleague who talks to customers.
What only you can supply, and what AI does well
Most advice on AI copywriting for websites is about prompts. The bigger decision is which parts of the job to hand over at all.
The human vs AI content question matters less than it seems. Google says its systems focus on the quality of content rather than how it was produced, and its spam policy on scaled content abuse targets pages generated in bulk mainly to manipulate rankings, however they were made. Does Google penalise AI-generated content? walks through the guidance. Buyers ask a simpler question: is this page specific, accurate and useful to me?
What only your business can supply
- The questions customers really ask. They are in enquiry emails, sales calls and the things your team explains every week.
- Prices and process. What things cost or what changes the price, what happens after someone gets in touch, and who they will deal with.
- Proof. Projects you can show, clients who have agreed to be named, certifications you actually hold, photos of your own work.
- Opinions. When you would advise against something, which option you recommend and why. Expertise shows most clearly in what you are willing to say no to.
- First-hand detail. The mistake new clients often make, the question that catches people out, the part of the job that takes longest. Google’s helpful-content guidance asks whether content clearly demonstrates first-hand expertise, and E-E-A-T explained covers how to make yours visible.
What AI does well, and where it doesn’t
AI earns its place on form: structure, alternatives, tightening and consistency, such as spotting that one page says “consultation” and another “discovery call”. It fails wherever a page needs facts it wasn’t given.
| Task | Hand it to AI? | Watch for |
|---|---|---|
| Turning your notes into a first draft | Yes | Gaps filled with plausible guesses |
| Headline and button alternatives | Yes | Ten versions of the same idea |
| Shortening a long section | Yes | The one specific fact getting cut |
| Writing a page from a one-line prompt | No | Copy any competitor could publish |
| Prices, timelines and guarantees | No | Numbers that sound right but aren’t yours |
| Projects, results and testimonials | No | Invented outcomes and quotes |
| Legal, safety or regulated claims | No | Confident statements nobody checked |
A six-step workflow for AI website copy
Deciding who writes which page, and by when, belongs in your website content plan. This is how each page gets written.
1. Gather source material before you open the tool
Collect what a good copywriter would ask you for:
- Enquiry emails, and the replies that won the work
- Notes or transcripts from sales calls, with the customer’s permission
- Proposals and quotes, including what you exclude
- Reviews and feedback, in customers’ own words
- The queries in Google Search Console’s Performance report that already bring visitors
- A recorded chat with whoever knows the service best: “what do people get wrong about this?”
Privacy first. Remove names, contact details and anything confidential before pasting material into an AI tool, and check your plan’s data settings, including whether your inputs can be used for training.
2. Brief each page
A brief turns raw material into instructions. Half a page is enough:
- The reader: who lands on this page, and what they already know.
- The page’s job: the one action you want next.
- The questions it must answer, taken from your source material.
- The facts it can use: prices, process and proof, written down.
- Words to use and avoid: your names for things, and phrases that aren’t you.
- Length and place: roughly how long, and which pages link to it.
A brief replaces the model’s assumptions with yours, and it works just as well for a human writer. For the blocks a service page needs, see how to write a service page.
3. Outline around buyer questions
Decide the headings before any drafting. Write them as the questions a buyer asks, in the order they ask them. For an office fit-out company’s main service page, an illustrative outline might be:
- What do you handle, and what do we still do ourselves?
- Can we keep working while it happens?
- What does it cost, and what changes the price?
- What happens if something goes wrong or runs late?
- What have you done for businesses like ours?
AI makes a useful sparring partner here: ask it what a sceptical buyer would still want to know, then keep only what matches real customer questions. Question-led headings with a direct answer underneath also make it easier for search engines and AI assistants to see what each section covers.
Then note which facts each section needs and whether you have them. A gap found now takes a phone call to fill; found after design approval, it holds up the launch.
4. Draft from your material, not the model’s memory
Now let the tool write, one section at a time for longer pages. Give it the brief, outline and source material, and set rules:
Write the section “What does it cost, and what changes the price?” for the page in the brief. Use only facts from the source material below. Where a fact is missing, write [NEED: …] instead of guessing. Keep customers’ own phrases where they fit. No superlatives and no claims we can’t prove. About 150 words, then two shorter alternatives for the first sentence.
The [NEED: …] markers matter most. They turn the model’s habit of filling gaps into a to-do list for you.
Match your brand voice. AI tools tend to pick up tone better from examples than from adjectives. Instead of “friendly but professional”, paste two or three passages you’re proud of and a short list of words you use and never use.
5. Verify every claim
AI website content fails most often on facts. At the time of writing (July 2026), even the strongest tools still state wrong things fluently, so check every draft against your sources:
Delete what you can’t verify rather than softening it. In regulated fields such as health, finance, property or law, have whoever is responsible for compliance read the page too.
6. Edit for voice, then test with someone who talks to customers
Read the page aloud. Wherever you stumble, or hear something you would never say to a customer, rewrite it in your own words. Cut sentences that summarise the paragraph before them.
Then hand it to a colleague who deals with customers every day, such as the salesperson or whoever answers the phone. Ask them:
- Is anything wrong or out of date?
- Would a customer understand every term?
- What would a customer ask next that the page doesn’t answer?
Their answers are often the most valuable edit the page gets. Then read it once more on a phone.
Tell-tale signs of generic AI copy
Generic copy follows patterns. Search your draft for these and rewrite each one:
| Sign | Typical example | Write instead |
|---|---|---|
| An opener that states the obvious | “Choosing the right partner is crucial for your business.” | The reader’s problem, or your answer, in the first line |
| Adjectives with no evidence | “Reliable, experienced, customer-focused service” | One fact that proves one of them |
| Abstract filler | “Your journey”, “tailored solutions”, “navigating complexity” | What actually happens, in plain words |
| Hedged claims | “Can help improve efficiency” | What changes, for whom, and how you know |
| Everything in threes | Every section ends with three parallel benefits | As many points as are true, of different lengths |
| “Whether you’re X or Y” | “Whether you’re a start-up or an enterprise, we’ve got you covered” | Who the service is really for, and who it isn’t |
| Vague “it depends” | “Pricing varies based on your needs.” | The three or four things the price depends on |
The swap test. Replace your business name with a competitor’s. If every sentence still works, the page isn’t yours yet. Add something they couldn’t say: a step in your process, a price range, a real project, an opinion.
Other languages: write for each audience, not word for word
If your customers use more than one language, it’s tempting to finish the page in one and have AI translate the rest. Machine translation is often good at literal meaning. Persuasion is where it falls short:
- People search differently. Search phrases in one language are often not translations of those in another, so a literal version can miss the searches that matter.
- Tone and formality don’t transfer. How directly to address a buyer, and how formal a sales page should sound, varies between languages and cultures.
- Idioms and examples break. A reference that lands in one language can confuse, or offend, in another.
- The errors are invisible. If you don’t read the language, you can’t tell a fluent mistranslation from a good page.
A better workflow for AI translation for websites:
- Brief each language from the source material and page brief, not from the finished page in another language.
- Keep a glossary of service names, product terms and how your brand name is written.
- Let AI draft in the target language if it helps, then have a native speaker who knows your field edit and approve it.
- Keep the facts identical (prices, terms, guarantees) and let idiom, examples and emphasis change.
- Check each version in the real layout on a phone, where longer words and headings can break buttons and menus.
Forms, error messages and confirmation emails need the same care. How to plan a multilingual website covers ownership and review, and the multilingual SEO guide covers how each version gets found.
What to do next
Start with one page: the service page that brings in most of your enquiries, or should. Run the swap test, gather the source material and work through the six steps. You’ll come out with a brief template, voice examples and a glossary to reuse on every page after it.
For campaign pages, how to write landing page copy applies the same research-first approach to headlines and objections, and how AI is changing websites shows where AI fits across the rest of your site.
If you’re planning a new website, content is often what holds a project up, so start early. Our website design and development projects begin with a strategy session and sitemap, so every page has a job before anyone writes a word. Once the sitemap is agreed, you get a clear list of what each page needs. Bring your source material to the first conversation: whoever writes the final words, it’s what makes the pages yours.