Artificial intelligence in advertising

The practical effect of generative tools on advertising is a large reduction in the cost of producing material. That is significant and it is narrower than the discussion suggests.

What has genuinely changed

The bottom four rows are where advertising is actually decided, and they are unaffected.

Where it earns its place

Volume tasks. Fifty product descriptions, thirty ad variants, translation into four languages. Work that was previously not done because it was uneconomic.

First drafts. Getting from a blank page to something to react to. The draft is usually mediocre and reacting to it is faster than starting cold.

Analysis. Summarising customer feedback, categorising enquiries, finding patterns in review text. Genuinely useful and previously skipped.

Concepting. Producing rough visuals to communicate an idea before committing to production.

Where it fails predictably

Anything requiring current facts, which it will invent confidently. Anything requiring knowledge of your business, prices, capacity or actual capabilities. Anything requiring judgement about tone in a sensitive context. And anything where being wrong is expensive.

The process that works

1. A specific brief, as detailed as one given to a person.
2. Generate, several variants rather than one.
3. Select, which is a human judgement and the actual value-adding step.
4. Rewrite, because the selected draft is a draft.
5. Verify every fact, figure, name and claim.
6. Name the person responsible for the published version.

Step five is where organisations get into trouble. Published material carries the same liability regardless of how it was produced.

The homogenisation risk

Everyone using similar tools with similar prompts produces similar output. The recognisable house style, the specific turn of phrase, the particular way a business talks, all erode.

The counter is a documented tone, real examples, and human editing that actively removes the generic register. Otherwise the cost saving buys indistinguishability.

The disclosure question

Requirements are emerging in several jurisdictions for labelling synthetic media, particularly images and video of people. The rules are moving and the direction is towards more disclosure, not less.

The practical position: keep a record of what was generated and how, and label synthetic imagery of people. That record is cheap to maintain now and expensive to reconstruct later.