Using AI in a marketing workflow without losing control

In this guide

Where AI earns its keep in a marketing week (and where it doesn’t)
Rule one: AI drafts, a named person approves
Data protection: what you can and can’t paste in
Ad claims and brand voice: the CAP Code doesn’t care who wrote it
Platform AI features: the settings to check
AI content and Google Search: volume without value
A sign-off checklist
FAQ

AI can draft a lot of marketing work, but the advertising rules apply “in the same way, regardless of whether AI is used or not” (ASA). So my rule is simple: AI and automation produce drafts and flags, and a named person checks them before anything goes to a client, an ad account or a website. Below is where that works, what to keep out of prompts, which platform AI settings to check, and a sign-off checklist.

Where AI earns its keep in a marketing week (and where it doesn’t)

AI earns its keep on work that repeats every week and has a clear right answer to check against. It’s a poor fit wherever a wrong answer could be published or relied on without anyone looking.

Heavy use is still rare. The ONS estimates 30% of UK businesses used at least one AI technology in September 2026, rising to 38% of businesses with 10 or more employees (ONS, BICS Wave 165). These are official statistics in development, so treat them as a guide. Only 10% of firms with 10 or more employees report extensive use (ONS).

In my week, automation and AI tools handle the repetitive work:

Checks across 31 Google Ads accounts every weekday.
A daily flag for manual CPC bids above a £5 threshold.
A weekly search term report listing terms that spent without converting, with suggested negatives.
Monthly GA4 reports for 36 websites, drafted automatically and then checked by a person before sending.

Each of these gives a person something to review before anything changes or goes out. If reporting is where you’d like to start, the one-page GA4 report people actually read covers what to put in front of people.

Don’t let AI work alone on anything relied on as fact. The ICO’s view is that the accuracy you need depends on the purpose: output relied on as fact needs more accuracy than output used for “inspiration”, and accurate training data doesn’t stop hallucinations (ICO).

Rule one: AI drafts, a named person approves

Give every AI or automated output one person whose name is on it.

Accountability and governance is one of the five principles in the UK government’s AI regulation white paper (GOV.UK, last updated August 2023). It works just as well as a rule inside a marketing team.

In practice:

A named owner per output type. Someone owns the monthly reports, someone owns ad copy, someone owns bid changes. If nobody’s name is on it, it doesn’t go out.
Dry runs before changes. My monthly bid adjustments (device, age, gender, location, schedule) are set from 180 days of results, with a dry run first.
Review suggested negatives before adding them. The reviewer knows things the data doesn’t, such as a product line that’s about to launch.

The same idea runs through five Google Ads automations that pay for themselves: the tool flags and a person decides.

Data protection: what you can and can’t paste in

Paste as little personal data into an AI tool as you can. Ideally none.

The ICO’s “Guidance on AI and data protection” is organised around accountability, transparency, lawfulness, accuracy, fairness, security and data minimisation, and individual rights, and includes an AI risk toolkit (ICO). One caveat: it was last updated on 15 March 2023, and the page says it’s under review because of the Data (Use and Access) Act and “may be subject to change”.

Data minimisation is the principle to keep in mind when prompting. The CRM data I work with includes customer names and contact details. None of it needs to be in a prompt.

The rules I’d suggest (general recommendations, and no substitute for legal advice):

Strip identifiers first. Remove names, email addresses and postcodes before pasting anything in.
Use business tools with training switched off. Keep client and customer data out of consumer AI tools.
Keep a list of approved tools. If a tool isn’t on the list, it doesn’t get client data.

Ad claims and brand voice: the CAP Code doesn’t care who wrote it

If AI writes an ad, you’re still responsible for every claim in it.

The CAP Code says ads “must not materially mislead or be likely to do so” (rule 3.1), and you must hold documentary evidence for claims before the ad runs (rule 3.7) (CAP Code section 3). CAP’s May 2025 guidance says the same rules apply whether AI is used or not, and an AI label rarely cures a misleading ad (ASA). On disclosure, it suggests asking whether the audience would be misled without it, and whether the disclosure clarifies the ad or contradicts it.

Platform-generated text is where this catches people out. Google’s text customisation writes extra headlines and descriptions from your landing pages, ads and keywords, and it’s on you to keep your site content accurate (Google Ads Help). An old price left on a landing page can end up in an ad you never wrote.

Brand voice is the other check, because AI drafts can sound generic. Start from headlines that already performed and edit from there, as set out in writing ads from what already worked.

Platform AI features: the settings to check

Platform defaults switch AI features on, so audit them every time you build a campaign and after every platform update. The short version, with sources below:

Feature	Default	Where to check
Google AI Max for Search	On for new Search campaigns	Brand and URL controls
Google text customisation and final URL expansion	On when an existing campaign opts in	“Added by” column, asset report
Meta “adapt to placement”	On	Advantage+ creative settings
Meta text improvements, generated backgrounds, image expansion, music	Set one by one	Paused ad preview

Google Ads: AI Max and text customisation

AI Max for Search is a layer of settings inside a Search campaign, on by default for new Search campaigns. When an existing campaign opts in, text customisation and final URL expansion are on by default, and turning text customisation off also turns off URL expansion. The controls are brand and URL inclusions and exclusions (Google Ads Help).

Text customisation (previously automatically created assets) refreshes at least every 48 hours. Review what it wrote in the “Added by” column and the asset report. From September 2026, campaigns using text customisation are automatically upgraded to AI Max (Google Ads Help), so check which of your campaigns now run it.

Meta: Advantage+ creative

Since Marketing API v22.0, Meta has deprecated the “standard enhancements” bundle, so each Advantage+ creative feature is opted in or out separately. “Adapt to placement” is on by default, and ads using AI-generated features need to be created paused and previewed first (Meta for Developers). That preview is the moment to check generated text and images against your claims and brand.

AI content and Google Search: volume without value

Google’s guidance says generating many pages with AI that add no value can breach the scaled content abuse policy (Google Search Central).

The same page asks you to fact-check AI output before publishing, including the parts readers don’t always see: titles, meta descriptions, structured data and alt text. Merchant Center requires labelling of AI-generated product images and text (Google Search Central, updated 1 October 2026).

A sign-off checklist

Before any AI-assisted output goes live, the named owner checks:

Owner: one named person is responsible for this output.
Purpose: if it will be relied on as fact, check it against the source.
Data: no names, emails or postcodes went into the prompt, and the tool is on the approved list.
Claims: every ad claim has documentary evidence on file before it runs (CAP Code 3.7).
Landing pages: prices, offers and dates on the site are current, because Google can turn them into headlines.
Google Ads: AI Max, text customisation, final URL expansion, brand and URL controls, and the asset report reviewed.
Meta: each Advantage+ creative feature set on purpose, and the ad previewed while paused.
Web content: titles, meta descriptions, structured data and alt text fact-checked.
Changes: bid or budget changes run as a dry run first.

FAQ

Do I have to label AI-generated ads in the UK?

CAP’s guidance asks whether the audience would be misled without a disclosure, and says a label rarely cures a misleading ad (ASA). Google Merchant Center does require labels on AI-generated product images and text (Google Search Central).

Does Google penalise AI-generated content?

Google’s guidance says generating many AI pages that add no value can breach its scaled content abuse policy, and asks you to fact-check AI output before publishing (Google Search Central).

Can I paste customer data into an AI tool?

Keep it to a minimum, in line with data minimisation in the ICO’s AI guidance, which is under review and may change (ICO). Strip names, emails and postcodes first, and use approved business tools with training switched off.

For the wider picture, see my guide to AI in marketing.

11 sources, all opened and checked

Disclosure of AI in advertising: striking the balance between creativity and responsibility, ASA and CAP
CAP Code section 3: Misleading advertising, ASA and CAP
Guidance on AI and data protection (under review), Information Commissioner’s Office
Accuracy of training data and model outputs (generative AI consultation response), Information Commissioner’s Office
A pro-innovation approach to AI regulation: white paper, GOV.UK
Using generative AI content, Google Search Central
About AI Max for Search campaigns, Google Ads Help
About text customisation, Google Ads Help
Advantage+ creative: get started, Meta for Developers
Business insights and impact on the UK economy: 8 October 2026, Office for National Statistics
Artificial intelligence in UK businesses: 2023 to 2026, Office for National Statistics