How to Choose a Marketing Agency That Uses AI Responsibly

Choose an agency that will put its AI use in writing: which tools touch your account, what a person checks before anything is published, who owns the work, how your customer data stays out of model training, and how synthetic images are labelled. Ask to see a real workflow and an AI clause in the contract before you sign.

Don't look for an agency that avoids AI. Almost all of them use it somewhere, and a good one passes some of the time saved on to you. Look for one that can explain where it uses AI and where it won't, because anything published under your name, from a claim about your services to a photo of a wedding, is your reputation, whoever or whatever produced it.

Follow me on Instagram@sagnikteaches

The example here is a wedding planner: a three-person business planning around 30 weddings a year, looking for an agency to run Instagram, a monthly newsletter, the website's journal pages and a modest ads budget. Weddings make the risks vivid, because couples buy on trust and on pictures, and AI can fake both.

Connect on LinkedInSagnik Bhattacharya

Where an agency's AI touches a small client's account

Start by knowing which parts of the work are likely to involve AI, so you can ask about each one. The table sets out the common uses and what careful handling looks like.

Subscribe on YouTube@codingliquids

A few platform facts are worth knowing because they catch careless agencies out. Instagram has capped hashtags at five per post since December 2025, so a caption with fifteen is a sign nobody is keeping up. On Meta, Advantage+ is the default for sales and leads campaigns, meaning the platform's own AI makes many targeting choices; a responsible agency will tell you what it still controls and how it checks the spend. And Google's guidance on AI content says generative AI is fine when it's accurate and useful, but that using it to produce many pages without adding value for users can breach its spam policy on scaled content abuse.

Seven questions that separate careful agencies from careless ones

Ask every agency the same seven questions, ideally in writing so you can compare the answers side by side.

1. "Which AI tools will you use on our account, and on which plans?" You want named tools on business terms, not "a mix of the best AI". Business plans from the major providers don't train on your content by default; personal accounts rely on each user switching training off.
2. "Who reads and approves work before it goes live?" A named person on their side, plus your sign-off for anything factual about your services, prices or events.
3. "Where won't you use AI?" The best answer draws lines: no fake testimonials, no synthetic photos presented as real events, no AI replies to complaints without a person.
4. "Who owns what you make for us?" The contract should assign the work to you. Note that the AI providers' own terms, such as OpenAI's and Anthropic's, pass you their rights in outputs "if any", so ownership of heavily AI-generated material can be less solid than it sounds. The tutorial on whether you own the AI-generated content you publish explains why.
5. "How do you handle images from AI tools?" Listen for rights awareness. Adobe says Firefly's own models are trained on licensed content, but partner models inside the Firefly app aren't covered by that, and its intellectual property indemnity is for enterprise customers only.
6. "What customer data will your tools see?" Ideally none beyond what's necessary, and never your full client list pasted into a chatbot.
7. "How will you tell us if your AI use changes?" A promise of notice before new tools touch your account or data.

Careful agencies answer these quickly, often with a policy document they already have. Careless ones answer with enthusiasm about output volume. For the brand-voice side of this conversation, how agencies keep brand voice consistent in AI-written content shows what a good process looks like from inside the agency.

Synthetic images and the weddings that never happened

Here is the risk most specific to businesses that sell experiences. AI image tools can now produce a convincing photo of a candlelit reception in a barn that never hosted one, with guests who don't exist. For a wedding planner, posting that as "our work" is misleading couples about what you have actually delivered, and if it's found out, it undermines every real photo you've ever posted.

A responsible agency will set rules like these, and you should expect to see them in writing:

• Real events are shown with real photos, credited to the photographer, with permission.
• AI may be used for mood boards, backgrounds on graphics and concept visuals, labelled as concepts ("styling concept, not a real event").
• Retouching stays within normal limits: removing a fire exit sign, yes; adding a string quartet, no.
• Synthetic images keep any AI metadata the tool adds. For online shops this is a formal requirement: Google's guidance says AI-generated product images must carry the IPTC metadata that marks them as made by a trained algorithm.

If you sell to customers in the EU, the AI Act's transparency duties, which have applied since 2 August 2026, include labelling deepfake content, so realistic synthetic images of people and places need care. The tutorial on labelling AI-generated images and video on social media covers the practical labels platforms offer.

Reviews, testimonials and engagement: lines no agency should cross

Some agencies sell "reputation" packages that stray into fake engagement, and AI makes fake reviews cheap to write. Google's review rules are strict: no incentives for reviews, no review gating (only asking happy customers), and, since April 2026, no staff quotas for reviews and no asking customers to mention specific things such as a staff member's name. Any agency proposing AI-written testimonials, "review generation", or bulk-bought followers should be off your list, whatever else it offers.

Replies to reviews are a legitimate place for AI drafts, with a person approving. An illustrative draft reply to a two-star review of a wedding shows why approval matters:

"We're so sorry your day didn't meet expectations! We'd love to make it right. Please contact us for a full refund of your planning fee."

The draft is polite and fast, and it has just offered a full refund that nobody authorised. The approved version kept the apology, removed the refund, and invited the couple to a call with the planner herself.

Give them a test brief in the pitch meeting

Slides describe a process; a small live task shows it. Give each shortlisted agency the same brief and ask them to talk you through how they'd produce it, including where AI comes in. The wedding planner used this one:

Test brief (please bring to the pitch meeting)

We're launching a micro-wedding package: up to 30 guests,
weekday dates only, starting at $6,500.

Please draft three Instagram posts announcing it, and show us:
- which parts you'd draft with AI and which you'd write yourselves;
- what you'd need from us (photos, facts, approvals);
- what you'd check before anything is published.

One agency returned drafts that looked polished but contained three problems. This excerpt is illustrative:

"90% of couples now prefer intimate weddings! Our micro-wedding package gives you the magic of a big day for less, with guaranteed perfect weather on our partner terraces. #microwedding #intimatewedding #weddingplanner #elopement #weddinginspo #bride2027 #weddingday #love #luxurywedding #weddingideas"

The statistic is invented, "guaranteed perfect weather" is a promise nobody can keep, and there are ten hashtags where Instagram now allows five. None of that would matter much if the agency's own process had caught it; the fact that it arrived at a pitch meeting unchecked is the real finding. The two better agencies brought drafts with no statistics, a list of questions about which terraces were actually available on weekdays, and a note that they'd use the planner's own photos from two small weddings she'd run.

Contract wording to add before you sign

Most agency contracts say little or nothing about AI. Ask for a short clause that turns the pitch-meeting answers into commitments. The version below is a starting point to adapt with your own adviser, not legal advice; the tutorial on AI clauses for agency contracts discusses each element in more depth.

Use of AI tools
1. The Agency may use AI tools to assist with drafting, editing and
   analysis. It will use only business-tier tools that do not use
   Client materials to train AI models, and will list those tools
   on request.
2. A named member of the Agency's staff will review all content
   before it is sent to the Client for approval. No content will be
   published without the Client's written approval.
3. The Agency will not use AI to create testimonials, reviews or
   images presented as real events, people or results.
4. Any synthetic or substantially AI-generated image will be
   identified to the Client before approval and labelled where
   platform rules or law require.
5. Client customer data will not be entered into AI tools except as
   agreed in writing for a stated purpose.
6. All deliverables are assigned to the Client on payment, to the
   extent the Agency holds rights in them.
7. The Agency will give 14 days' notice before introducing a new AI
   tool that processes Client data.

What AI should do to the price you're quoted

Agencies price in different ways: a monthly retainer for a bundle of deliverables, a day rate, or a fee per piece. AI has cut the drafting time for many of those pieces, so it's fair to ask where that saving has gone. There are three honest answers: more work for the same fee, more time on strategy and checking, or a lower price. The answer to worry about is none of those, where the same hours are billed as before while the drafting now takes a fraction of the time and nobody extra is checking.

Ask each agency to split its quote into four parts: strategy and planning, creating content, review and approvals, and reporting and meetings. You're not trying to haggle over minutes; you're checking that the review work exists. An illustrative split from a careful agency on an $1,800 monthly retainer might be 30% strategy and ads management, 35% content, 20% review and approvals, and 15% reporting. If a proposal allocates nothing to review, the AI drafts are coming straight to you, and your evenings become the quality control.

A quick sum shows what volume pricing can hide. Agency A's $1,200 for 20 articles works out at $60 an article, even before you count the daily posts. A useful article about a real wedding needs an interview with you or the couple, photos used with permission and a careful edit. Ask how many hours of a person's time go into each $60 article; the answer tells you whether anyone is checking, and whether the articles can be about your real work at all.

Worked example: a wedding planner compares three agencies

The planner shortlisted three agencies and received these illustrative proposals. The figures are what these particular agencies quoted, not a guide to market rates.

Agency A offered the most content for the money, and that was the problem: twenty articles a month about weddings, for a planner with 30 weddings a year, could only be produced at volume by AI with little real detail, exactly the pattern Google's guidance warns about. Agency C was affordable and willing, but thin on process. Agency B cost the most per month, but its answers held up at every step.

The planner didn't simply pay the higher fee. She asked Agency B for a lighter version: two articles a month instead of four, with the ads management kept. B quoted $1,450. Over a year that is $17,400, against $10,800 for Agency C. She judged the $6,600 difference worth it for the review process and the policy she could hold them to, but she also set a checkpoint after 60 days, which is where the next section comes in.

The first 60 days: spot checks that keep an agency honest

A good start can drift once the agency is busy. A monthly half hour of checking keeps standards where they were promised.

• Sample five published posts. Are the facts right? Five hashtags or fewer? Real photos for real events?
• Read one article closely. Does it include detail only you could have supplied, or could it describe any planner anywhere?
• Ask for the month's AI-use note. A short list of which tools were used on which pieces. Agencies with a policy can produce this easily.
• Check review replies. Any promise, discount or admission should have come to you first.
• Look at the ads report against the raw numbers. If the summary says bookings rose, check the enquiries in your own inbox.

Checking the AI-written monthly report against your inbox

Many agencies now generate the narrative part of their monthly reports with AI, and a summary can be confidently wrong about the numbers underneath it. This illustrative paragraph arrived in the planner's second monthly report:

"Instagram ads drove a 38% increase in website conversions this month, showing strong demand for the micro-wedding package."

Her own enquiry inbox told a smaller story: micro-wedding enquiries had gone from four to five. The 38% came from the ad platform's conversion count, which included visits to any page, and it compared a partial first month with a full second one. Two platform details make this kind of misreading more likely. Since March 2026, Meta has counted click-through attribution on link clicks only, which changes how results compare with older reports, and platform AI such as Advantage+ reports on outcomes it helped choose. A responsible agency explains changes like these rather than letting a headline number stand alone.

The corrected paragraph read: "Ads sent 212 visitors to the micro-wedding page. Five enquiries came through the website form this month, up from four, and two mentioned Instagram. It's too early to judge the package; we suggest keeping the budget flat for another month." That is less exciting and far more useful, and it's the standard to ask for from the first report onwards.

Decide in advance how you'll tell customers about AI in your marketing, if at all; should you tell customers when you use AI in your marketing sets out the options. And if part of the reason you're hiring an agency is search visibility in AI answers, the companion tutorial on choosing an SEO agency that understands AI search covers the questions specific to that work.

Agencies and AI: common follow-up questions

Should an agency tell me when it has used AI on my work?

It should tell you how it uses AI in general, in writing, and flag anything unusual, such as a fully synthetic image or an AI-written article with little human editing. You don't need a note on every caption. What matters is that a named person reviewed the work and that nothing misleading about your business goes out under your name.

Will AI-written content hurt my search rankings?

Not because it's AI-written. Google's guidance says generative AI content is acceptable when it's accurate and useful, but that producing many pages without adding value for users can breach its spam policy on scaled content abuse. An agency offering dozens of thin articles a month for a small business is the risk, whatever wrote them.

Can an agency put my customer list into AI tools?

Only if you've agreed to it and the tools are on business terms that don't train on your data, ideally covered by a data processing agreement. Ask which tools would see customer names or emails and why they need to. Segmenting a list rarely needs personal data pasted into a chatbot.

Who is responsible if an AI-made post or image breaks the rules?

Usually you, as the business publishing it, at least in the eyes of your customers and the platforms, although your contract can require the agency to cover its own mistakes. That's why the approval step and the contract wording matter. For anything legal, such as advertising rules in your sector, ask a qualified adviser.

Further reads

• AI Marketing Tools vs Hiring a Marketer: A Cost Comparison — Compare the agency route with tools plus your own hire.
• How to Build a Brand Voice Guide That AI Can Follow — Give any agency a voice guide its AI can actually follow.
• How to Check AI Content for Plagiarism Before You Publish — A quick check you can run on the agency's drafts.
• Social Media Pre-Publish Checklist for AI-Drafted Posts — The approval checklist to use on AI-drafted posts.
• How to Write an AI Disclosure Statement for Your Website — Publish your own short statement on how you use AI.
• How to Hire a Part-Time Marketing Assistant Who Uses AI — When a part-time hire makes more sense than an agency.
• Is It Cheaper to Use AI or a Freelancer for Marketing Content? — A per-piece cost comparison of AI and freelance writers for marketing content, counting your own editing time, with three worked cases.
• Questions to Ask Before You Pay Anyone to Set Up AI Marketing — 30 questions in six groups, with what a good answer sounds like, plus a filled-in comparison of two real-looking quotes for a food truck.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Search Central guidance on using generative AI content (updated 10 December 2025); the series fact sheet for Instagram hashtag limits, Meta Advantage+ defaults, Google review rules, Adobe Firefly terms and EU AI Act Article 50.