Automation and AI

Generative AI

Also called gen AI

Models that produce new text, images, audio or code from a prompt, rather than only sorting or scoring existing data.

Quick facts: Generative AI

Category
Automation and AI
Also called
gen AI
Level
Beginner
Affects
Content production speed, editing workload, brand voice
Where to see it
ChatGPT, Claude, Gemini, Copilot, assistants built into marketing tools
In this article4
  1. How generative AI works
  2. Why generative AI matters
  3. Common mistakes with generative AI
  4. How to act on it

How generative AI works

These models are trained on very large collections of text, images or audio, and what they learn is pattern: which word tends to follow which, which shapes and colours tend to go together. When you give one a prompt, it produces the continuation that fits the patterns best. Nothing is retrieved from a filing cabinet; the output is composed on the spot, which is why the same prompt twice can give two different answers.

That is the difference from the older kind of marketing AI. Predictive systems score things that already exist: which of these leads is likely to buy, which bid is likely to win. Generative systems make something new. The two are often bundled into the same product, and it is worth knowing which one you are relying on for a given task.

The important consequence is that the model has no concept of true, only of likely. Fluency and accuracy are separate properties, and it is very good at one of them.

Why generative AI matters

For a small marketing team, the honest gain is on the parts of the job that are slow rather than difficult. First drafts, restructuring something you already wrote, turning a long page into a short one, producing several headline variants to test, writing alt text, translating, summarising a pile of customer feedback, drafting the tracking code snippet you would otherwise search for. The blank page stops being the bottleneck.

It also lowers the cost of doing the boring work properly. Descriptions for a large product catalogue, subject line variants, or a first pass at reformatting an old post are all jobs that used to be skipped for lack of hours.

Common mistakes with generative AI

Publishing unedited output is the obvious one, and it shows. Generated copy tends towards a flat, agreeable middle: correct in shape, empty of anything only your business could have said. Search engines reward first-hand experience and expertise, and a page assembled from patterns has neither.

The subtler mistake is asking it for facts. Prices, specifications, statistics, quotations and sources are exactly where a fluent model will produce something plausible and wrong. Using it to shape what you already know is safe; using it to find out what you do not know is not.

How to act on it

Give it the raw material rather than the topic. Your notes from a client call, your own rough answer, the transcript of the question a customer actually asked — output improves enormously when the model has something specific to work from instead of a subject line.

Keep a fixed rule about facts: anything checkable gets checked against the source before it is published, and anything you cannot verify comes out. Read every line aloud once; if it sounds like it could belong to any business in your industry, it needs your own examples in it. Understanding the model behind these tools makes both habits feel less like caution and more like common sense.

Do and do not

Do

  • Use it for drafts, structure and variants, not facts
  • Feed it your own notes, calls and customer questions
  • Check every name, claim and source before publishing

Do not

  • Publish anything you have not read line by line
  • Ask it for statistics or citations and trust them
  • Let it write in a voice that is not yours

Questions people ask about this

Will Google penalise content written with AI?

Google's stated position is that it judges the content, not the tool that produced it. Helpful, original material written with assistance is fine; thin pages produced at scale to fill a site are not, whoever or whatever wrote them. The practical risk is not a penalty but irrelevance: generic pages that add nothing tend not to earn rankings in the first place.

What should I never use generative AI for?

Anything where being confidently wrong is expensive. Legal, medical and financial claims, prices and specifications, statistics, quotations attributed to real people, and case study results all need a verified source. Use it to draft, restructure and summarise material you already trust, and keep it away from work where the fact itself is the deliverable.

How do I stop AI-written copy sounding generic?

Feed it something only you have. Real customer objections, the way your team actually explains a service, a specific job you completed, the constraints of your market. Then edit hard: cut the opening sentence, cut anything that would be true of a competitor, and add a detail that could only come from your business. Generic input produces generic output.

Related terms

Found this useful?

Share it, or ask an AI to summarise it

Back to the glossary

Knowing the term is the easy part

Applying it to your own site and budget is the work. Book a call and I will tell you what actually applies to you.