The short version

Writing website copy with AI works best as a five-step process, not one magic prompt. Collect a brief, decide the message for each section, and let AI draft. Then rewrite with your own facts, publish, and measure. Skip those steps and the output is fluent but sounds like everyone else. The model is not the main problem. Most prompts start without a brief, a brand voice, or a review step. You remain the final editor.

This article explains how to set up that process, where AI helps, and where human judgment is still needed. It covers chat tools such as ChatGPT and Claude. Then it looks at an AI agent such as Hermes, which can remember context, reuse proven steps, and run recurring jobs.

Why AI copy often sounds generic

A language model does not know your business. Without context, it fills the gaps with common patterns: corporate filler, superlatives, and a neutral tone that belongs to no one. The grammar may be fine, but the copy still feels artificial because it makes no distinctive choices. Google’s guidance on generative AI content points the same way. It warns against generating many pages without adding value for users, and it asks you to fact-check AI output before you publish.

The answer is not a cleverer prompt that asks the model to sound “more human.” Give the model the information it naturally lacks:

  • Who you are: your business, your offer and what customers come to you for.
  • How you sound: two or three of your own example sentences from texts you like.
  • What the goal is: the page, its intended reader and the action you want them to take.
  • What is not allowed: claims you can’t prove and words that don’t fit your brand.

That brief gives the model something specific to work with. It turns a loose prompt into a repeatable process.

What AI does well now (and where it falls short)

AI is strongest between the rough idea and the first draft. It can structure, rewrite, and shorten copy, produce alternatives, adapt a draft for a particular audience, and keep repetitive product descriptions consistent. It can also draft metadata and product description variations, provided you supply the facts.

How editing changes the copy: an example

Suppose a landscaping company asks AI to draft a services page. The first version says: “We deliver top quality and work closely with every customer to make their dream garden a reality.” It is grammatically correct but empty. Any competitor could publish it unchanged.

For illustration, here is a version with fictional business details. Replace them with verified facts before you publish: “Our four-person team builds your garden in two to three weeks, using our own machinery and plants from nurseries around Aalst. You also receive a maintenance plan for the first year, because the work does not stop when the planting is done.” It fills the same space but now gives readers concrete reasons to choose this company.

AI cannot decide which details matter most to your customers. That is why source material makes such a difference. Ask it to “write something about garden landscaping” and you get the first version. Provide the company’s process, region, and real promises, and the draft has a chance of becoming the second.

AI is weak wherever real knowledge of the business matters. Think of how you work, the results you have achieved, and the claims you can prove. Humor that suits your audience and the choice of which message goes on which page also need you. The tool can only speed up the drafting.

Writing website copy with AI in five steps

This workflow suits homepages, service pages, product pages, and landing pages. Blog articles need extra work around search intent, structure, and measurement. The guide to blogging with AI covers that workflow.

Step 01

Collect the goal, reader, and voice

For each page, note its purpose, target audience, desired action, and two or three examples from your own writing that set the tone. Put it in a short brief you can reuse. Here is a template, filled in for the fictional landscaping company from the example above:

Page: services page for garden landscaping
Reader: homeowners near Aalst who want a new garden, not maintenance
Goal: explain what we build and how long it takes
Action: book a site visit
Voice examples: two sentences from our own newsletter
Facts we can prove: team size, own machinery, local nurseries, maintenance plan
Words we avoid: top quality, dream garden, passion

The fictional details stay fictional until you swap in your own. The last two lines matter most: they tell the model what it may claim and which clichés to leave out.

Step 02

Decide the message for each section

Choose which points the page needs to make and in what order. Keyword research shows which words your readers use. AI can help brainstorm, but the structure is your decision.

Step 03

Ask AI for a first draft

Give the tool your brief and structure. Ask for two versions of each section instead of one long draft. That way you can edit more precisely. A prompt can be as simple as this:

Use the brief below.
Write two short versions of the opening section of a services page.
Use only the facts in the brief. Do not invent numbers.
Do not use the words listed under Words we avoid.

Step 04

Rewrite with your own facts

Replace generic statements with your process, examples and provable claims. Read the result aloud to check that it still sounds like your business.

Step 05

Publish and measure

Add accurate metadata, publish the page, and check Google Search Console after at least 30 days for impressions and clicks.

The fifth step is easy to forget, but it separates using AI from finding out whether the copy works. A page nobody finds was an expensive prompt. The guide to setting up Google Search Console explains how to set up that measurement.

Writing in a chat versus working with an AI agent

ChatGPT and Claude in the browser are good starting points. They are accessible, capable writers and useful for experiments. Their limitation is the chat format: each new session can mean starting from scratch. You may need to paste the brief, examples and brand rules again, and refinements from earlier chats are easy to lose.

An AI agent can retain more of that process. I work with Hermes Agent, and three features make it different from an isolated chat:

  • Memory: the agent can retain your brand voice, approved brief, and earlier copy. Later iterations build on what you have already recorded.
  • Skills: a proven approach can become a reusable procedure. The editing process lives in the tooling rather than in a list you paste into every chat.
  • Scheduled jobs: recurring work such as reviewing copy, refreshing seasonal pages and fetching measurements can run on a schedule.

For a one-off page, a chat tool is often enough. If you write or update pages every month, an agent that retains the process can save more time and keep the work consistent.

How Straffe Sites uses AI for website copy

At Straffe Sites, copy does not run from prompt to publication automatically. Hermes agents prepare the work; people decide. Here is a practical example:

The agent works from the brief, delivers a complete landing page package, and marks the steps that need human review. The publishing checklist of this run lists two. First, replace all fictional example details with verified facts before publication. Second, have the owner check the claims about crew size, duration, machinery, nurseries, and the maintenance plan.

A person still decides which claims are valid, whether the tone is right, and what gets published. That is how I use AI for copy: it speeds up preparation and rewriting, while the final edit remains human. Prefer to start from a generated design rather than written text? That route follows the same split between agent preparation and human decisions: from AI design to a working website.

Common mistakes with AI copy

  • Publishing without editing: the first version is a draft. Without your own facts, it may sound polished but say very little.
  • One prompt for the whole website: every page has its own goal and reader. A generic briefing makes all pages the same.
  • Leaning on superlatives: “top quality” and “personal approach” appear on millions of sites. Replace them with concrete details about how you work.
  • Ignoring search intent: even elegant copy will struggle if it does not match the query. Add accurate metadata and measure the result.
  • Failing to record the brand voice: without examples and agreed rules, every session becomes another guess. Document the tone once and reuse it.

Conclusion

Writing copy with AI is an editing process, not a prompt trick. It needs a brief based on your own material, a structure you choose, a faster first draft, and a human final edit. ChatGPT and Claude work well for occasional drafts. For recurring work, an agent with memory, reusable skills, and scheduled jobs can retain more of the process. Measuring and preparation can run on that same schedule. See how AI automation for small businesses handles it with clear rules. Prefer not to manage the briefing and editing yourself? Straffe Sites builds websites with a clear content structure and provides copywriting and SEO that reflect your business and how customers search. Where AI assistants should cite your copy, our GEO approach picks up from there.