I've spent the last two months using these tools inside real client engagements rather than testing them on toy prompts, and I've formed strong opinions about where they earn their keep. The dividing line isn't the one most of the commentary assumes — it isn't 'long content bad, short content good'. It's about whether the task requires information the model could not possibly have.
Where they're genuinely excellent
Transformation tasks. You have something good and you need it in another shape: a webinar transcript into a blog outline, a long case study into five LinkedIn posts, a technical explanation rewritten for a non-technical buyer. The source material carries the value; the model handles the mechanics. This is where I've seen the largest real time savings, and the quality is genuinely good because the substance came from a human.
Where they cost you more than they save
Anything requiring a position. Ask for an opinionated piece on your category and you'll get the consensus view, elegantly expressed, because consensus is what it was trained on. It reads well, which is the dangerous part — a bad idea in fluent prose survives review far longer than a bad idea in clumsy prose. I've watched teams ship AI-drafted positioning that said nothing, precisely because nobody noticed it said nothing.
- Excellent: reformatting, summarising, first drafts of structured content, variant generation
- Adequate with heavy editing: explainers on well-documented topics
- Poor: positioning, opinion, anything that must contradict the consensus
- Actively risky: statistics, case studies, anything factual it might invent
The fabrication problem is not theoretical
Ask for supporting statistics and you will frequently receive plausible, well-formatted, entirely invented numbers with confident attributions to real organisations. This is the single biggest risk in using these tools for B2B content, because your credibility is the product. A prospect who checks one figure and finds it doesn't exist will discount everything else you've published. My rule in client work is absolute: no number goes out unless I've personally seen the source.
A workflow that actually holds up
Human decides the argument and the angle. Human supplies the raw material — the interview, the data, the customer quote. Model produces structure and a first draft. Human rewrites the parts that carry the opinion, which is usually the opening, the transitions and the conclusion. Human verifies every factual claim. It's still faster than writing from nothing, and what ships is genuinely yours.
Use them for the mechanics and keep the judgement. The moment you outsource the thinking, you've automated your way to sounding exactly like everyone else in your category.
Want to apply this to your business?
Book a free 30-minute strategy call and we'll diagnose your specific growth blockers together.
Book a Free Call