Mastering AI tools for content creation

I use AI every working day. I also implemented the AI chatbot that now handles first-line queries on our WooCommerce store. So I am not writing this as a sceptic, I am writing it because most advice on AI content is either breathless or dismissive, and neither matches what actually happens when you put these tools into a real pipeline.

Here is the honest version: AI is excellent at the middle of the process and poor at both ends.

Where it earns its place

Think of a content pipeline as five stages: decide → research → draft → edit → publish.

• Decide: what to write, for whom, why now. AI is bad at this. It has no access to last week's sales meeting or the fact that your best customer keeps asking the same question.
• Research: genuinely strong. Summarising, structuring, finding the counter-argument you have not considered.
• Draft: strong, with a caveat below.
• Edit: strong for mechanics, weak for judgement. It will tighten a sentence beautifully and cheerfully remove the one specific detail that made the paragraph worth reading.
• Publish: irrelevant. Metadata, scheduling and distribution are your job.

Most people get poor results because they hand over stages one and five and keep stages two through four. Do the exact opposite.

The brief is the whole job

A vague prompt produces the average of everything ever written on the topic, which is precisely the content nobody needs more of. A specific brief produces something usable.

Mine always contains six things:

1. Audience: not "marketers" but "a founder with 2,000 SKUs who has never run a technical SEO audit".
2. The one thing they should do differently after reading.
3. Three real specifics: a number, a tool name, a thing that actually happened. This is what stops the output reading like everyone else's.
4. Structure: how many sections, roughly what each covers.
5. Voice constraints: mine are: British spelling, no bullet lists longer than five, never open with a rhetorical question, never use "delve", "leverage" or "in today's fast-paced world".
6. What to leave out: usually the most valuable line in the brief.

The verification problem

This is the part people skip, and it is the part that will eventually cost someone their job.

Language models produce fluent, confident, well-formatted claims that are sometimes simply false. Statistics, dates, quotes and citations are the highest-risk categories, because they are exactly the things that look most authoritative and are least likely to be checked.

My rule is blunt: every number, name and date in published content gets traced to a primary source, or it comes out. Not "sounds plausible". Traced. If the model cites a report, I open the report. Roughly one in five citations I check does not say what the draft claims it says.

The same applies to anything that touches regulated ground. In healthcare e-commerce (which I have worked in) an invented claim is not an embarrassment, it is a compliance incident.

Where AI genuinely changed my week

Not blog posts. The unglamorous volume work.

• Product descriptions at catalogue scale. With 2,000+ SKUs, the choice was never "AI or a copywriter", it was "AI or nothing at all on 1,700 of them". Structured input, strict template, human review on the top 200 by revenue.
• Metadata. Title tags and meta descriptions to a character budget, in bulk. Mechanical, rule-based, perfectly suited to it.
• First-line customer queries. The chatbot handles the twenty questions that make up most of the volume, and hands off cleanly rather than guessing.
• Repurposing. One article into a LinkedIn post, an email and five social captions. This is transformation, not creation, and it is where the quality holds up best.
• Adversarial review. "Argue against this strategy as a sceptical finance director." Genuinely useful before a budget meeting.

The homogenisation risk

If every brand in your category uses the same three models with the same lazy prompts, every brand in your category starts sounding identical. That is already visible on LinkedIn, and search engines are getting better at spotting it.

The defence is not to avoid AI. It is to feed it things it cannot know: your own data, your own numbers, your customers' actual words from support tickets and reviews, your own opinions. The differentiator was never the writing. It was the specifics.

A workflow you can copy

1. Pick the topic from a real customer question. Never from a keyword tool alone.
2. Write the six-point brief by hand. Ten minutes.
3. Ask for an outline first. Fix the outline. Ninety per cent of bad drafts are bad outlines.
4. Generate section by section, not all at once. Quality degrades over long outputs.
5. Rewrite the opening and closing yourself, always. They carry the voice.
6. Verify every factual claim against a primary source.
7. Add one thing only you could have written, a number from your account, a mistake you made, a client conversation.
8. Read it aloud. If it sounds like a press release, it is not finished.

Step seven is the whole ballgame. Everything else is typing.

---

← Back to the journal