Nyah Eliud
The honest position
We use AI writing tools every day and we aren’t going to pretend otherwise. We also never publish what they produce without a person rewriting it. That isn’t caution for its own sake. These tools fail in specific, predictable ways that do real damage, and the damage is hard to spot if you aren’t looking for it.
The useful question isn’t whether to use them. It’s which parts of the job they’re actually good at.
What it is genuinely good at
- Structure. Turning a messy brief into an outline, and spotting the sections you forgot.
- First drafts of things you already know. You can fix a mediocre draft far faster than you can face a blank page.
- Repetitive work with a clear pattern. Product descriptions, meta descriptions, alt text, ten versions of an ad headline.
- Reading. Summarising a long document, comparing competitor pages, pulling the questions out of a pile of customer emails.
- Translation and tone. Moving between French and English, or making something formal read plainly.
Notice what those have in common. You already know whether the answer is right. That’s the condition that makes these tools safe.
Where it fails, expensively
It invents specifics. Prices, dates, statistics, regulations and names. Confidently, in the same voice it uses for things that are true. A page quoting a made-up figure about your own industry does more damage than no page at all.
It doesn’t know your business. It will write that you’re passionate about customer service, because that’s what such pages say. Everything that really makes you different has to come from you.
It doesn’t know this market. Ask about Cameroon and you get text that sounds plausible, reads generic, and is sometimes wrong. Mobile money, how people search here, and what things really cost aren’t things to take on trust.
It heads straight for the average. These models produce the most typical version of a piece of writing. In a market where your competitors are using the same tools, that’s a plan for sounding exactly like everyone else.
The rule we useNever let it state a fact you can’t check yourself. Ideas and structure from the machine. Every number, claim and promise from a human who knows.
What Google actually says about it
Google’s position is narrower than either the panic or the hype suggests. It doesn’t punish content for being written by AI. It punishes unhelpful content made mainly to rank, whoever or whatever wrote it.
In practice that means a thin, generic article churned out at scale is a risk no matter who wrote it, and a genuinely useful page is fine no matter how the first draft appeared. The tool isn’t the problem. Publishing without judgement is.
How we actually use it
Research and competitor reading first. Then an outline, argued with and rewritten. Then a draft that a person restructures, cuts by a third, and fills with the specifics only we know: what we charge, what we’ve seen go wrong, what happened on a real project. The two assistants we keep open, and what each is better at, are on the tools we use.
The result is faster than writing from nothing and better than either of us could manage alone. What it isn’t is automatic.
Good and bad uses, side by side
| Task | Use AI? | Why |
|---|---|---|
| First draft of a service page from your notes | Yes | Saves time; you supply every fact |
| Shortening a long email or proposal | Yes | It is good at compression |
| Social captions from a photo and a few words | Yes, then edit | Fast, but check the tone |
| Prices, timescales, guarantees | No | It will invent plausible numbers |
| Case studies and testimonials | No | They must describe what really happened |
| Fifty pages to rank for every town | No | Thin, repetitive pages are what Google calls spam |
| Replies to a complaint | Draft only | A person must read and own every word |
Google’s position is that it rewards helpful, people-first content however it is produced, and treats content made mainly to manipulate rankings as spam whoever wrote it. The test is not who typed it, but whether it is useful, accurate and specific.
A workflow you can copy
Give it the audience, the one question the page answers, and five things you know that a competitor couldn’t have written. Ask for an outline. Fix the outline yourself. Then ask for one section at a time.
Then read it out loud. Anything you would be embarrassed to say to a customer in person, cut. That step alone removes most of what makes AI writing easy to spot.
Where this fits with the rest of the work is covered in getting your business mentioned by ChatGPT and Gemini, and our automation service handles the parts that should really run themselves.
Go further: the AI and automation guide, our automation service, automation prices, or the five automations to build first.
Questions people ask
Not for being AI-written. Google's guidance rewards helpful content however it is made, and acts against content produced mainly to manipulate rankings, whether a person or a machine wrote it.
Where readers would reasonably want to know, yes. What matters most is that a person checked every fact and stands behind the content, which is how we work.
It can draft them from your specifications, which saves time. Check every size, material and price against your own records before publishing, because it will fill gaps with guesses.
Comments
No comments yet.Be the first to add one. We read every comment and reply to the ones that ask something real.
Leave a comment
Thank you. It will appear once we have read it, usually the same working day.