How Agencies Use AI: What Clients Should Ask
Home  /  Blog

How Agencies Use AI: What Clients Should Ask

The Short Answer: Most agencies use AI to draft copy, sort keywords, summarise reports and run bidding inside Google and Meta. That is fine. What matters is which tasks use it, who checks the output, what happens to your customer data, and whether the time saved goes back into your account. Ask those four things in writing before you sign, and again at every quarterly review.

Almost every agency now says it is “AI-powered”. The phrase tells you very little. One agency uses AI to write a first draft that a senior writer then rebuilds. Another pastes your brief into a chatbot and sends back whatever comes out. Both can honestly say they use AI, and the results for your business will look nothing alike.

This guide from IZI Digital Marketing looks at how agencies use AI from the client’s side of the table. It maps where AI usually sits in the work, which questions separate careful agencies from careless ones, what Malaysia’s data rules mean for your customer lists, and where the time savings should go. If you are still building a shortlist, our guide to choosing a digital marketing agency in Malaysia covers the wider decision. The short video below explains why a human still needs to stay in the loop.

Actionable AI for Marketers: The Human in the Loop

Source video: Watch on YouTube

PART 1 · DIAGNOSE

How Do Marketing Agencies Actually Use AI?

IN BRIEFAgencies use AI in two very different places. Some of it is tooling the agency chooses, such as writing assistants and research tools. The rest is built into the ad platforms themselves, such as automated bidding. You can shape the first through your contract. The second needs sound settings and good data, which is why it belongs in any digital marketing RFP.

When clients hear that agencies use AI, they often picture one tool. In practice it is a stack of small uses spread across the whole service. Sorting them into three groups makes the questions easier to ask:

Where the AI sits Typical examples Who controls it
Agency’s own tools Drafting blog posts and ad copy, keyword grouping, image variations, meeting notes The agency, so you can set rules in the contract
Ad platform automation Smart Bidding, Performance Max, AI Max for Search, Meta Advantage+ Google or Meta, steered by the agency’s settings and your data
Analysis and reporting Report summaries, anomaly alerts, competitor scans, search term reviews The agency, with your data flowing through the tool

The second group is the one clients forget. Google describes Performance Max as a campaign type that uses Google AI for bidding, targeting and creative. Meta’s Advantage+ suite does the same on Facebook and Instagram. Here the agency’s skill is in the inputs: conversion tracking, audience signals and the assets it feeds the machine. Our guide on when Performance Max works and when it doesn’t shows how much depends on those inputs.

Bottom Line: Ask about AI in all three places. An agency that only talks about its writing tools may be ignoring the platform automation that spends most of your budget.

Not sure how much of your current account is run by automation?

The Blueprint’s Diagnose phase maps which parts of your marketing are human-led and which are machine-led, so you know what you are actually paying for. See how the Blueprint works

BENCHMARK BRIEFING 1 OF 4

Which Agency Tasks Are Most Likely to Use AI?

IN BRIEFBid management and first drafts sit at the top of the list, because the platforms automate bidding by default and writing tools are cheap. Strategy, client advice and final approval sit at the bottom. The more a task shapes your spend or your brand voice, the more human review it needs, which your agency account manager should be able to explain.

Share of Work on Common Agency Tasks That Is Typically AI-Assisted (%)
Illustrative estimate of the share of work on six common digital marketing agency tasks that is typically AI-assisted in 2026, from bid management to strategy and client advice.
Agency task Typical AI-assisted share
Bid management in Google and Meta

85%

First drafts of ad copy and blog posts

70%

Keyword research and grouping

60%

Report summaries and commentary

45%

Design and image production

35%

Strategy, budgeting and client advice

10%

Illustrative model by IZI Digital Marketing, built on the default automation in Google Ads and Meta Ads and common tool use in Malaysian marketing agencies, 2026. Shares are indicative judgement, not a measured survey, and vary widely between agencies.

Most agencies use AI heavily at the top of this chart, so read it from the bottom up. The tasks with the least AI are the ones you are really paying a consultant for: deciding where your money goes and why. If an agency’s pitch leans heavily on how much it automates, ask what its people spend their saved hours on. A good answer names planning, testing and closer reading of your results. A weak answer is simply “more clients”.

PART 2 · DESIGN

What Should You Ask an Agency About AI?

IN BRIEFYou do not need a list of 30 questions. Five well-chosen ones cover the real risks: which tasks use AI, who reviews the output, where your data goes, who owns the work, and where the saved time goes. Put the answers in writing and check them during marketing agency due diligence.

Most question lists online are written by agencies selling AI services, so they test how advanced the tools are. From the client’s chair, the useful test is whether the answers are specific. Here is what to ask, and what a good or worrying reply sounds like:

Ask this A good answer A worrying answer
Which of my deliverables use AI? A task-by-task list, naming the tools “We use our proprietary AI across everything”
Who checks AI output before it goes live? A named role, a checklist, and what they change “The tool is very accurate”
Does my customer data go into any AI tool? Which data, which tool, and whether it is used for training “Don’t worry, it’s all secure”
Who owns AI-assisted content and creative? You do, stated in the contract No clause, or ownership “depends on the tool”
Where does the time saved by AI go? More testing, more analysis, or a wider scope for the same fee A vague reply, or a subject change
Consultant’s Note: Ask one follow-up that most clients skip: “Show me one piece of work before and after your review.” A careful agency can pull up an AI draft next to the published version in a minute. The gap between the two tells you more than any policy document, because it shows what a human actually added.

Add one more question if you work in a crowded market: does the agency use the same AI prompts, templates or trained models for your competitors? That links to the wider issue of agency exclusivity and serving your rivals, which matters more once your brand voice and data sit inside shared tools.

Bottom Line: Judge the answers on how specific they are. Vague replies about “secure, proprietary AI” are the clearest warning sign you will get.

BENCHMARK BRIEFING 2 OF 4

Where Does AI Help an Agency, and Where Does It Hurt?

IN BRIEFWhen agencies use AI, it saves the most time on repetitive work with a clear right answer, such as grouping keywords. It carries the most risk where a mistake is public or expensive, such as claims in ads or health advice in blog posts. The review effort should follow the risk, not the time saved, as with AI content versus human writers.

Time Saved vs Error Risk When Agencies Use AI, by Task
Illustrative comparison of typical time saved, error risk and the level of human review needed when a marketing agency uses AI for five tasks: keyword grouping, report summaries, ad copy, blog content and customer data analysis.
Task Typical time saved Error risk Human review needed
Keyword grouping 50–70% Low Spot check
Report summaries 40–60% Medium Check every number
Ad copy variations 30–50% Medium to high Every line, for claims and policy
Blog and web content 20–40% High Full edit plus fact check
Customer data analysis 30–50% High (privacy) Approved tools only, anonymised data

Illustrative model by IZI Digital Marketing, built on common agency workflows and published platform and search guidance, 2026. Time-saved ranges are indicative judgement, not a measured survey.

Notice that blog content saves the least time once it is done properly. A publishable article still needs a human who knows the subject, adds real experience and checks every fact. Google’s own guidance on generative AI content warns that producing many pages without adding value can breach its spam policies. So an agency promising “four times the content for the same fee” is often promising more pages, not better ones. For SEO services, fewer, stronger pages usually win.

PART 3 · DEPLOY

Is Your Data Safe When an Agency Uses AI Tools?

IN BRIEFIt depends on which data goes into which tool. Aggregated campaign figures carry little risk. Customer names, phone numbers and enquiry details are personal data under Malaysia’s PDPA, and your business stays responsible for them. Agree the rules in writing, the same way you would for your PDPA website compliance.

Malaysia tightened its data rules recently. The Personal Data Protection (Amendment) Act 2024 renamed data users as data controllers and brought in duties such as data breach notification. The Commissioner has also issued guidelines on cross-border transfer of personal data. Many AI tools process data on overseas servers, so pasting a customer list into a chatbot can be a cross-border transfer. Use this box to decide what is acceptable:

DECISION BOX · WHICH DATA CAN YOUR AGENCY PUT INTO AI TOOLS?

Type of data Free consumer AI tool Business AI tool with no-training terms
Public info (your website, published ads) Acceptable Acceptable
Aggregated campaign results Avoid if it reveals margins or targets Acceptable
Customer lists, leads, enquiry messages Not acceptable Only anonymised, and only with your written approval
Sensitive data (health, finance, children) Not acceptable Not acceptable without legal advice

Verdict: Default to “no personal data in AI tools” and allow exceptions in writing. It is easier to loosen a strict rule than to recover leaked customer data.

Write these rules into your agency scope of work or a short AI clause, not just an email. This article is general guidance, not legal advice; for sensitive sectors, check the PDPA position with a lawyer.

Bottom Line: Your business carries the PDPA duty even when the agency makes the mistake. Set the data rules before any tool touches your customer list.

BENCHMARK BRIEFING 3 OF 4

How Long Does It Take for Agency AI to Pay Off?

IN BRIEFNot in the first month. Platform automation needs a learning period and steady conversion data, and agency tools need tuning to your brand voice. Expect a flat or slightly worse start, then gains from around the third month if tracking is clean, as with Smart Bidding strategies.

Cost per Lead Over Six Months After Switching to AI-Led Campaigns (Index, Month 0 = 100)
Illustrative time series of indexed cost per lead from month 0 to month 6 after a switch to AI-led bidding and creative, comparing an account with clean conversion tracking against one with weak or broken tracking.
Month Clean conversion tracking Weak or broken tracking
Month 0 (before switch) 100 100
Month 1 108 115
Month 2 100 118
Month 3 92 116
Month 4 87 112
Month 6 82 110

Illustrative model by IZI Digital Marketing, built on the learning periods and data requirements described in Google Ads and Meta Ads documentation, 2026. Index values show a pattern, not measured client data. Lower is better.

The gap between the two lines comes down to what the AI is told to chase. If tracking counts page views or duplicate form fills as “conversions”, the system gets very good at finding the wrong people. Before you agree to more automation on Google Ads management or Meta, ask the agency to show which actions count as a conversion and how they were tested.

Worried your campaigns are optimising for the wrong conversions?

Share how your tracking is set up and we will help you judge whether the automation has a fair chance of working. Ask for a tracking second opinion

PART 4 · DRIVE

Should AI Make Your Agency Cheaper?

IN BRIEFNot always, and a lower fee is not the only fair outcome. AI cuts hours on routine tasks, but adds new work in review, testing and tool costs. The fair test is whether you get more value for the same spend. Build that into your marketing agency SLA and reviews.

Clients often assume that if agencies use AI to write first drafts, fees should fall. The picture is more mixed. Agencies now pay for tool licences, spend more time checking output, and run more tests because variations are cheaper to make. What you should expect is a visible return on the saved time, in one of these forms:

  • More testing: more ad variations, landing page tests or audience experiments each month, with results reported.
  • Deeper analysis: time moved from building reports to explaining them and recommending changes.
  • Wider scope: an extra channel, language version or content type within the same retainer.
  • Faster turnaround: shorter delivery times written into the agreement, not just promised in the pitch.

Ask the agency which of these you are getting. Our guide on PMax vs Meta Advantage+ shows how much judgement is still needed even when the platforms automate most of the buying, and Meta Ads management follows the same logic.

Bottom Line: Do not measure AI’s value by the fee alone. Ask where the saved hours went, and expect a specific answer each quarter.

BENCHMARK BRIEFING 4 OF 4

Where Do the Hours Saved by Agency AI Go?

IN BRIEFIt depends on the agency model. A consulting-led team tends to put saved hours back into testing and analysis on your account. A volume-driven agency tends to use them to take on more clients. Neither is hidden, if you ask. The answer is a useful filter when comparing a Malaysian marketing agency shortlist.

How Agencies Typically Reinvest Hours Saved by AI, by Agency Model (%)
Illustrative split of how hours saved by AI are reinvested at three types of marketing agency: into testing and analysis on the client account, into extra scope for the client, or into taking on more clients.
Agency model Testing and analysis / extra scope / more clients (%)
Consulting-led or boutique

55 / 25 / 20

Mid-sized full-service agency

35 / 25 / 40

High-volume package agency

15 / 15 / 70

Illustrative model by IZI Digital Marketing, built on typical service models and staffing patterns at Malaysian marketing agencies, 2026. Bar key: ink = testing and analysis on your account, rust = extra scope for you, orange = more clients per team. Shares are indicative, not a measured survey.

A high-volume agency using AI to serve more clients is not doing anything wrong. It keeps packages affordable. The problem is only a mismatch between the model and your needs. If your account is complex, with several channels and a long sales cycle, you need saved hours to come back to you. Ask how many accounts each team member now handles compared with two years ago.

THE VERDICT

Judge the Agency’s Judgement, Not Its Tools

How agencies use AI will keep changing, and the tools named in any pitch will be out of date within a year. What stays useful is a short set of questions: which tasks use AI, who reviews it, what happens to your data, who owns the output, and where the saved time goes. An agency that answers all five specifically, in writing, is showing the judgement you are paying for.

Start with those questions in your next brief or quarterly review. If you want AI search itself on the agenda, our guide on getting your brand cited in AI Overviews covers that side. For the wider hiring decision, return to the digital marketing agency Malaysia guide.

FAQ

Frequently Asked Questions

1. Is it bad if my marketing agency uses AI?

No. It depends on how the agency uses it, but AI for drafts, keyword work and bidding is now normal. The risk comes from unchecked output, careless data handling, or AI replacing the thinking you pay for.

2. Should my agency tell me when it uses AI?

Yes. How much detail you need depends on your industry, but you should know which deliverables use AI and who reviews them. Ask for this in writing, and include it in your contract or scope of work.

3. Does Google penalise AI-written content?

Not for being AI-written. It depends on quality: Google says it rewards helpful, original content however it is made, but mass-producing pages without added value can breach its spam policies. Human editing and real expertise still matter.

4. Can my agency put my customer data into ChatGPT?

Not without your approval. It depends on the tool’s terms and the data type, but customer details are personal data under the PDPA. Agree which tools are allowed, and keep personal data out of free consumer tools.

5. Who owns content my agency creates with AI?

You should. The answer depends on your contract more than the tool, so make sure it states that all deliverables, including AI-assisted copy and images, belong to your business once paid for.

Want to know how much of your marketing is really human-led?

A free Blueprint consultation helps you decide which AI use in your account to keep, question or rewrite into your contract, based on how your campaigns are actually run.

Book my free consultation

Have a campaign in mind? Let's talk.