"AI digital marketing" is a phrase covering at least a dozen separate jobs, sold as though it were one product. Some of those jobs are genuinely solved and you should be using AI for them today. Some are barely started. And a few get measurably worse when automated, which nobody selling them mentions. This sorts the whole field, with the parts that are actually working separated from the parts that are being marketed.
The map
Solved
Campaign construction, creative production, keyword research, copy variants, performance monitoring, lead alerting. Use AI for all of these now — doing them by hand produces no advantage.
Partly solved
Audience selection (depends entirely on implementation), lead scoring, budget allocation, SEO content. Works when built properly, fails silently when not.
Not solved
Knowing your customer, pricing, positioning, judging whether a claim is true, deciding what a customer is worth. These are not waiting on a better model.
Actively harmed
Continuous editing of live campaigns, auto-publishing without review, automated replies to customers. Automation makes each of these worse.
Advertising: the most mature area
This is where AI has moved furthest, because the work is mechanical and the feedback loop is fast. A tool can now take a website URL and produce a structured campaign — audience segments, ad copy, an image with readable text, the correct objective — and launch it into your own ad account through the official APIs.
Two things separate a real implementation from a demo, and both are checkable.
First, targeting. Meta exposes a catalogue of roughly 284 behaviours through its API, and returns the same catalogue regardless of the country requested — country-specific entries sit inside one shared list. A tool with a hardcoded shortlist silently discards everything outside it. We found this in our own product: a hand-maintained list of fifteen behaviours was discarding 161 of 177 selections for having no matching ID, so every campaign outside the market that list was written for ran without the signals that existed for it. No error, no report, no way to notice.
Second, keywords. Ask whether search terms come from live Google Keyword Planner data through the official API, or from a language model asked what is popular. The second is common and produces lists that read well and contain terms nobody searches.
Creative: solved, with one specific caveat
Image and video generation is genuinely good now. The caveat is text, and it explains why most AI-generated ads look wrong.
Image models cannot spell. They produce pixels that resemble letters, because they were trained on pictures rather than on typography. Ask for an ad with "30% OFF" and you get something that reads as "3O% QFF" at a glance and worse on inspection.
💡 The fix is to stop asking the image model to write. Generate the scene with the image model, then draw the headline, price and call to action over it as real text through a layout engine in real fonts. It is then actually text — correctly spelled, on brand, and legible. Choosing the text colour by measuring the brightness of the region it sits on matters too: a white headline over a bright sky is invisible, and that is how a lot of otherwise good creative fails.
Video has moved similarly — a presenter avatar delivering a script, optionally built from a photo of the business owner, with captions, translation and dubbing. The real constraints are encoding and duration rather than generation.
Copy: fast, and legally dangerous by default
Generating ad copy is trivial for a language model. Two caveats cost real money.
Translation is not localisation. Copy translated from English reads as translated, because English ad writing leans on compression and wordplay that does not survive. Give the model the offer and the audience and have it write in the target language from scratch.
And models produce exactly the copy patterns every advertising regulator restricts — superlatives, invented statistics, outcome guarantees, testimonial framing — because that is what persuasive ad copy looks like in their training data. The advertiser stays responsible everywhere, and "the AI wrote it" has no standing.
💡 One rule removes most of this: no claim, figure or superlative goes into an ad unless you could produce the evidence within an hour. Specific verifiable claims also outperform superlatives, so it costs you nothing in results.
Lead handling: the most underrated area
Most AI marketing tools stop at the click. For any business selling appointments, quotes or consultations, that is half the job and the other half decides whether the first half was worth paying for.
Capture
Native platform lead forms complete materially better than landing pages, because nothing has to load. On one campaign we watched 45 clicks produce only 28 page arrivals — nearly 40% never got there.
Score
Rating each lead on stated budget, timeline, prior visits and device, so you call the big job first. The budget test must run in your own account currency — a threshold in the wrong currency can be off by 85×, and budget is usually the heaviest factor.
Alert
Reaching a phone, not an inbox, the moment it lands. Response time decides conversion more reliably than almost anything else.
Where automation actively hurts
Worth stating clearly because it runs against the marketing.
Every edit to a live ad set restarts the platform's evaluation of it, so a campaign rewritten daily can spend its whole first week never getting far enough into evaluation to deliver. "Continuous AI optimisation" as a selling point deserves real scepticism.
Auto-publishing removes the only compliance review you had. And an automated reply that gets it wrong costs more than a slow human one — acknowledge automatically, answer personally.
How to sequence it for a small business
First
Lead alerting. Cheapest to set up, highest return, no downside. Do it before anything else.
Second
Campaign construction and creative. The labour disappears, the approval stays with you.
Third
Performance monitoring and reporting, so you can look without assembling.
Last, carefully
Automated budget movement, and only with hard bounds on how much can shift and a floor it cannot cross.
The one question to ask any vendor
What does this do without asking me, and can I turn it off? A tool that publishes advertising, moves budget or messages customers autonomously is making decisions you are accountable for. That might be exactly what you want — but it should be a choice you made, not a default you discovered afterwards.
And the second: does it work in ad accounts you own? If it runs ads from an account the vendor controls, your pixel history, audiences and spend record are not yours, and leaving costs you all of it.
The whole loop, not one step of it
Adyft builds the campaign, generates the creative, launches into your own Google and Meta accounts, and alerts you when a lead arrives. 14-day free trial, no card, 0% of ad spend.
Start free — no card needed