What I Got Wrong About AI-Generated Content in 2023

Libby Day • July 16, 2023

In 2023 I published a post on this site that was written entirely by ChatGPT, credited to "ChatGPT" as the author, as a kind of experiment. At the time it felt like a novelty. Three years on, it reads like a time capsule of exactly the wrong lesson.

The mistake wasn't using AI. It's that the post skipped the part that actually matters: a process for using it well.

What I'd say to 2023 me

AI-generated content isn't good or bad in itself, it's unfinished by default. A first draft from a model is exactly that, a first draft. The gap between "AI wrote this" and "this is worth publishing" is entirely down to what happens after the generation: fact-checking, structural editing, adding the specific expertise and examples only you have, and cutting the parts that are technically correct but say nothing.

I've written about this more directly since, the issue was never the prompts. It was the missing system for using AI output as a starting point rather than a finished one.

What's different in 2026

Search engines and AI models have both gotten considerably better at telling generic AI output from content with real expertise behind it. The posts on this site that perform best now are the ones with a specific point of view, drawn from actual client work, not the ones optimised to sound comprehensive. Volume of AI-generated content is no longer an advantage, if it ever was. Distinctiveness is.

The practical takeaway

Use AI to draft, structure, and speed up research. Don't use it to skip the judgement calls that only someone who actually knows the subject can make. That's true for a blog post and it's true for most of what AI gets used for in marketing right now.

If you want help building a content process that uses AI properly instead of leaning on it, get in touch.

Digital Articles + Stories

A torn opening in yellow paper reveals a white layer printed with “Black Friday.”
By Libby Day • November 19, 2026
Plugin updates, new scripts and quick tweaks are how sites break at peak. What to freeze, what to check first, and how to handle the exceptions.
Neon-lit close-up of a face beside a glowing sign reading “GOOD VIBES ONLY”
By Libby Day • November 5, 2026
A raw AI visibility count tells you that you appeared, not whether it helped. Seven questions to put the number in context before you report it.
Hooded person overlaid with colorful digital code and data graphics on a dark background
By Libby Day • October 22, 2026
AI visibility is easier to see than it was. Outcomes still aren't. What you can measure, what you can only estimate, and how to report it honestly.
Abstract diagonal rainbow light bands and white reflections across a dark background
By Libby Day • October 15, 2026
Clients are right to ask about AEO. Here's what has genuinely changed in AI search, what hasn't, and what to ask before you buy it.
Colourful abstract geometric patterns in bright neon light
By Libby Day • October 1, 2026
Notes from a Duda Vibe webinar: why experience decides what AI delivers, how to use it for admin, and why prompting well keeps you in control.
Gray 3D lattice sculptures with white text reading “CREATING IN THE DIGITAL AND PHYSICAL REALMS”
By Libby Day • September 24, 2026
Notes from an Edinburgh Futures Institute panel on research and AI, through a maker's lens: incentive, intent, agency and who answers for the tools.
Blue-lit glass cylinders with scrolling code on a purple background
By Libby Day • September 15, 2026
GPTBot and ClaudeBot found zero pages linked only through JavaScript in a new controlled test, and Google's own indexing crawler barely fared better.
Hand touching clothes on a rack, with a white shirt featuring red stars in the foreground
By Libby Day • September 1, 2026
I tested Google Stitch, Bolt.new, Lovable and Duda Vibe on a real charity build, then compared all four to building it by hand. Here's what actually won.
More Posts