Content
How to use AI for content creation without sounding like AI
A practical way to use AI for content creation with ChatGPT, Claude or Gemini while keeping your company's voice. It includes a quick check, using a competitor's name, that shows whether a draft sounds like you or like anyone.
Why AI-generated content sounds generic
A common belief is that AI content sounds generic because the tools write badly. In fact ChatGPT can turn out a long, tidy, error-free post on self-assessment for sole traders in two minutes, one that any accountancy practice in the country could publish unchanged. Using AI for content creation is quick. Whether the result sounds like your company depends on the material you give it.
The reason is technical. A language model, the program behind ChatGPT, Claude or Gemini, writes by picking the most likely next word, again and again. Give it nothing of yours and it returns something close to the average of thousands of texts on the same subject. The result is clean sentences a hundred firms could sign.
What is missing is what only you know. The question a client asked on the phone yesterday, the mistake you see every quarter, your view on a new rule. A Harvard Business School and BCG study of 758 consultants found that AI users produced over 40% higher quality work and finished 25% faster on suitable tasks. Used outside its strengths, AI made results worse.
In writing, those strengths are organising ideas, suggesting structure and producing first drafts. Experience, judgement and checked facts have to come from you. The method below splits the work that way.
AI for content creation: a 6-step method
Order matters. The more of your own material you put in at the start, the less you will need to fix at the end.
- 01
Let the AI interview you first
Before asking for any text, ask for questions. For example, "I'm writing for my clients about self-assessment for sole traders. Ask me five questions, one at a time, to draw out what I know that isn't online." Answer as you would on the phone to a client, without polishing. You will come away with stories, objections and nuances no prompt could have produced.
- 02
Give it a voice guide and real samples
A voice guide is a short document describing how your company writes. How you address readers, sentence length, words you use and words you avoid. Add two or three of your own texts you like, such as a well-written email or a post that did well. AI copies an example far better than an abstract brief like "friendly but professional".
- 03
Ask for an outline, then each section
A whole article in one go tends to come out flat. Ask for an outline, correct it, then draft section by section, pasting in the interview answers that belong to each one. You stay in control of every part and can redo one without touching the rest.
- 04
Edit for specifics
Look for sentences that could sit on any competitor's website and swap them for an example, a real deadline, a local landmark or a case from your team. Cut the introduction if it says nothing and open with the most useful point. Short and specific usually beats long and vague.
- 05
Read it aloud
Your ear catches what your eye skips. If you would never say a sentence to a client, rewrite it. If every sentence is the same length, the text sounds mechanical, so mix short ones with the odd longer one.
- 06
Check facts, names and sources
AI can make up figures, studies or quotes and sound completely sure of them. Check every fact against its original source and remove anything you cannot verify. Take extra care with prices, dates and legal references. Before publishing, ask a colleague to read it.
The competitor test: swap the name to catch generic lines
It is the quickest way to tell whether a draft sounds like you or like anyone. You only need the text and the highlighter in your editor.
- 01
Swap your name for a competitor's
Replace your company's name with a direct competitor's throughout the text. Then read every sentence again, slowly.
- 02
Highlight the sentences that still hold
If a sentence is still true with the other name, highlight it. "We offer a friendly, professional service" survives the swap. "If a document is missing from your return, we ring you before the end of the month" probably does not.
- 03
See where the highlights cluster
If most sentences are highlighted, the text is interchangeable and it is worth going back to the interview. If only a few are, and they sit in the opening or the ending, touching up those parts is enough.
- 04
Rewrite with your own material
For each highlighted sentence, look in your notes for a fact, a case or an opinion only your company could sign. If you find nothing, the sentence probably does not need to be there.
AI writing tells worth removing
None of these is a mistake on its own. When they pile up, readers think of a machine before they think of your company.
Stock phrases
"In an ever-changing landscape", "now more than ever", "it's not just a product, it's an experience". They fill space and carry no information. Delete them and see whether the sentence still works.
Introductions that only announce
Opening paragraphs that say what is coming or restate the question in the title. The reader came for the answer, so lead with it.
Overly symmetrical lists
Everything in threes, bullet points of identical length, adjectives in matching pairs. Real life is rarely that neat. Give each list as many points as you actually have.
Dashes and dramatic colons
A dash-wrapped aside in every paragraph and sentences that build to a reveal after a colon. Used now and then, both are fine. Repeated throughout, they become a recognisable signature.
Worn-out words
Elevate, empower, robust, cutting-edge, journey. Once a word turns up in every generated text, it stops carrying weight. Look for the plain verb that says what you mean.
Example
A Bristol estate agency rewrites its listings from the viewings
Here is an illustrative case. An estate agency in Bristol with five negotiators lists several properties a week, and ChatGPT writes the descriptions from the property sheet. Most of them open with "a bright, spacious home in a sought-after location", and buyers keep ringing to ask things the listing should have told them.
The branch manager changes the method. Before drafting, the negotiator who showed the property answers five questions from the AI. What you hear from the bedroom, when the sun reaches the living room, what the last family who viewed it asked, what is within walking distance. Those answers, plus the agency's voice guide and two old listings that worked well, become the material for the draft.
Then they run the competitor test. "Excellent transport links" gets highlighted and replaced with the bus stop on the corner and the walk to the primary school. Before the listing goes live, the negotiator checks floor area, council tax band and service charge against the property sheet.
To measure the change, the agency compares enquiries per listing over two or three months, and how many calls repeat questions the text already answered. It also notes how long each negotiator takes to prepare a description. If enquiries improve and the time stays reasonable, the method stays.
One voice across the whole team
When several people write with AI, each one ends up with a different tone. The sales rep sends stiff emails, social posts fill up with exclamation marks and the website reads like another company. To avoid that, the team shares three things besides the tool.
The first is the voice guide, one or two pages kept where everyone can find it. The second is the prompts that work best, saved in a custom assistant if your tool allows it, or in a shared document. The third is a review step: anything going on the website or to a client gets a human read first, and any correction that keeps coming up goes into the guide.
To start, pick a text your company publishes often, such as a listing, a product page or a newsletter. Write a one-page voice guide, do the AI interview, draft the text and run the competitor test. Compare the result with the last one you published.
Obinova is an AI agency based in Barcelona and Paris. Our training covers ChatGPT, Claude, Gemini and Microsoft Copilot and works from your company's real texts, on a dedicated video platform or in person. If you want to look at the whole content process, our AI audit and consulting maps it with your team and shows where AI fits.
Frequently asked questions
Does Google penalise AI-generated content?
Google has said it rewards helpful, high-quality content however it is produced. Its spam policies target pages generated at scale to manipulate rankings that offer little to readers. An article drafted with AI, then reviewed and enriched with your own experience, falls outside that category. A generic, unedited text will simply find it harder to stand out.
Can you tell if a text was written by AI?
Automatic detectors exist, but their results are unreliable. They can flag human writing as AI and miss generated text altogether. Readers, on the other hand, quickly notice stock phrases and a lack of concrete detail. Your effort is better spent making each text useful, accurate and clearly your company's own.
ChatGPT vs Claude vs Gemini: which is best for content?
All three, like Microsoft Copilot, write well in English, and each has strengths depending on the task. The result depends mostly on your method, meaning the material you supply, the samples you share and the final review. Choose a tool your company has approved and test it with the same brief and your own texts before deciding.
How long does it take to write an article with AI?
It depends on the topic and how much of your own material you have. The draft is the quickest part. The time goes into the opening interview, the editing and the fact-checking, which are exactly what give the text its value. The process can feel slower at first. The sensible approach is to time two or three texts and compare with how long they used to take.
What content should you not hand over to AI?
Opinion pieces, sensitive messages to clients and anything involving legal, tax or technical claims are best written by you, or reviewed very carefully. AI can help you organise ideas or polish the style. Responsibility for what you publish stays with your company, so every figure and claim needs checking before it goes out.
How much does AI content creation cost?
The tools are usually charged per user per month, at prices each provider sets and changes. Free versions are fine for testing, though they limit usage and offer weaker data terms. The main cost is time: the interview, the editing and the fact-checking. If your team needs training, we put together a quote after a first meeting.