I Let AI Write This Article in 2023. Here’s What It Got Wrong About Marketing

The original article began when I handed a piece about artificial intelligence over to artificial intelligence.

I had barely used AI at the time. It was more of a novelty than a working tool: something interesting to experiment with and write about, rather than something woven into my day. I asked Jasper to produce most of the article, then gave its final question to ChatGPT. I published the result with very little intervention.

Three and a half years later, AI is a vital part of my working day.

I use it to explore unfamiliar subjects, find and interrogate sources, challenge ideas, structure research, improve writing, analyse information, solve technical problems and help manage the machinery behind Marketing Made Clear. It is not an oracle, a replacement for expertise or a licence to stop thinking. It is an unusually capable assistant whose authority needs to be earned task by task.

That is a much more positive view of AI than I had in 2023. It is also a much more sophisticated one.

AI has come a long way. More importantly, so has my understanding of when to use it, what to give it and when not to trust it.

The Confident Answer That Never Happened

The 2023 article told readers:

There are two fairly substantial problems with that sentence.

Turing’s paper was published in 1950, not 1956. Turing himself died in 1954, making a new paper from him two years later ambitious even by the standards of artificial intelligence.

The year 1956 does matter. It was the year of the Dartmouth Summer Research Project on Artificial Intelligence, organised by John McCarthy and widely treated as the birth of AI as a recognised field of study. The AI had combined a real person, a real paper and a real milestone into one fluent, authoritative and entirely incorrect statement.

This is the most important lesson in the old article. The problem was not that the writing sounded obviously robotic. The problem was that it sounded perfectly plausible.

It Was Not a Bad Article. That Was the Danger

Most of the original piece was broadly reasonable. It described AI as a technology that could support learning, pattern recognition and automation. It identified uses in customer segmentation, recommendations, personalisation, data analysis and customer service. It raised questions about education, transparency, bias, privacy and accountability.

Some of that now looks impressively prescient. The marketing applications it listed have become normal features of modern platforms and workflows. Its warning that education providers would need to respond to AI was correct. So was its expectation that explainability and ethics would become more important as adoption grew.

But the article also had the familiar qualities of early generative AI copy:

It could explain that AI might personalise an email. It could not tell you whether doing so would genuinely help the customer, weaken the brand, create a privacy problem or simply produce a great deal more mediocre email.

That difference matters in marketing. Producing material is not the same as exercising judgement.

What Changed Between 2023 and 2026?

The original article appeared on 29 March 2023, just two weeks after GPT-4 was released. The modern generative AI boom was already under way, but it was still young. Many people were trying these tools for the first time and working out whether they were a revolution, a parlour trick or a more efficient way to write a disappointing LinkedIn post.

Since then, the technology has become faster, more capable and more useful across text, images, audio, video, data and software. AI systems can work across multiple types of information, use tools, search sources, analyse files and complete sequences of tasks that would have required several separate applications in 2023.

Adoption has shifted with it. The Office for National Statistics estimates that reported AI use among UK businesses with ten or more employees rose from about 12% in late 2023 to about 35% in 2026. Yet the same research suggests that adoption remains fairly shallow: only 10% of AI-using businesses said they used it extensively.

That feels consistent with the world I see. AI has moved from novelty to normality, but familiarity is not the same as maturity. Plenty of organisations have access to AI without yet having a clear view of where it creates value, where it introduces risk or who remains accountable for the result.

The tools have improved dramatically. They have not made human judgement obsolete.

How AI Became Part of My Working Day

I use AI most comfortably for exploration and assistance. It can help me turn a broad question into a research plan, explain a subject, compare competing theories, identify gaps, find potential sources and test whether I have understood something properly. It can challenge an argument, play devil’s advocate and expose assumptions I may have missed.

It is also useful in the practical work around communication and marketing: structuring an article, improving clarity, adapting material for a different audience, developing campaign concepts, analysing customer feedback, conducting SEO research and interpreting performance data.

Beyond writing, it can help organise spreadsheets, build formulas, diagnose technical problems, review code, test webpages and take some of the repetition out of administration.

That is a broad range of uses, but there is an important distinction running through all of them: helping me do the work is not the same as taking responsibility for the work.

The AI Delegation Threshold

The way I now think about AI is as a sliding scale of authority.

A low-risk, reversible task using public information can sit towards the bottom of the threshold. A decision affecting someone’s employment, health, money, legal position or personal data sits much higher.

This is not a neat division between things AI can and cannot do. It is a judgement about what AI should be allowed to do in a particular context.

Where I Draw the Line

I am comfortable using AI to explain legal or regulatory wording, identify possible compliance risks and help review privacy documentation. I would challenge its conclusions and check anything consequential. It can assist with a financial model, but I would not ask it to recommend a major investment or expenditure decision.

It can draft a policy or contract; it cannot provide the final assurance that the document is legally suitable. I may discuss a health concern with AI before seeking professional advice, but I would not treat it as a source of definitive medical advice. The same applies to legal and financial guidance.

I am also comfortable using AI with many forms of work and personal information where the system, purpose and safeguards are appropriate. That does not mean every piece of information belongs in every AI tool. Unpublished commercial plans, employee information, sensitive health data, credentials and payment details all require a higher threshold.

The Information Commissioner’s Office makes the same underlying point in more formal language: when AI uses personal data, data protection must be designed into the system rather than added afterwards.

What This Means for Marketers

Marketing is unusually exposed to both the strengths and weaknesses of AI. It contains an enormous amount of work that AI can accelerate: research, segmentation, ideation, analysis, copy, visual development, optimisation and reporting. It also relies on context, taste, empathy, evidence and commercial judgement—qualities that cannot be safely reduced to an attractive paragraph produced in seconds.

1. Fluency is not evidence

An AI answer can be clear, detailed and wrong at the same time. If a claim matters, find the source. If the source matters, read it. A list of references generated by AI is not the same thing as research.

2. Direction determines quality

Ask AI for “an article about AI in marketing” and it will usually produce an average of the material it has seen: competent, familiar and forgettable. Better work comes from a clear question, original source material, a defined audience and someone willing to make decisions about the argument.

3. Personal experience is not decoration

The old article contained almost nothing of me beyond the experiment itself. That made it interchangeable with thousands of other explanations of AI.

This rewrite has a point of view because my relationship with the technology has changed. I can compare what I thought in 2023 with how I actually work in 2026. AI can help articulate that experience, but it cannot manufacture it.

4. Human review is not one final spellcheck

Review should happen throughout the process: defining the question, choosing sources, testing the reasoning, checking the facts, making editorial decisions and taking responsibility for publication.

Correcting a few commas at the end does not turn an AI-generated argument into a human-directed one.

5. Delegation must match consequence

AI suggesting ten headline options is not the same as AI deciding which customer receives a loan, which candidate gets an interview or whether an employee is behaving suspiciously. The fact that the same technology can contribute to both tasks does not make the risks equivalent.

What About Students and Junior Marketers?

The original article asked whether AI would help students cheat and concluded that institutions would need measures to prevent it. That was not wrong, but it was too narrow.

The more useful question is how people learn when a machine can produce a passable answer before they have developed the knowledge required to judge it.

Used well, AI can be a patient tutor. It can explain an unfamiliar concept in different ways, generate practice questions, challenge an argument and give feedback. Used badly, it can remove the productive difficulty through which understanding develops.

That matters for junior marketers too. If AI always does the first piece of research, develops the first structure and writes the first draft, where does the person build their own judgement?

The answer is not to pretend the technology does not exist. It is to make the learning visible: ask people to explain their reasoning, inspect sources, defend decisions and reflect on where AI helped or hindered them.

Was the 2023 Article Written by AI? Is This One?

That distinction is more meaningful than pretending every piece of work is either “made by a human” or “made by AI”. Modern work is often collaborative. What matters is who supplied the judgement and who is accountable when something goes wrong.

In 2023, I saw AI mainly as something novel enough to write about. In 2026, I see it as a powerful working tool.

My view is more positive because the technology is genuinely more capable and because I have experienced the value it can add. It is more sophisticated because I now understand that usefulness and trust are not the same thing.

The old article told us something important, although not in the way it intended. A machine could produce fluent marketing copy, make a basic factual error and give the reader very little reason to notice.

AI can explore, assist, advise and sometimes act. The skill is knowing where to set the threshold.