What makes an AI marketing provider the right choice?

The right AI marketing provider shows you a working system, names the person who reviews your work, keeps everything it builds in your name, handles customer data within NZ or Australian privacy law, and puts price and exit terms in writing. What the provider will show you before you sign tells you the most.

Labels are a weak signal in this market. Gartner estimated in June 2025 that only about 130 of the thousands of vendors selling agentic AI are real, in a warning about "agent washing", and predicted over 40% of agentic AI projects will be cancelled by the end of 20271. The same caution applies to marketing providers that have added "AI" to their name.

If you are still deciding which kind of help you need, start with what AI marketing services include, or the overview of AI marketing services across New Zealand and Australia. This guide assumes you have a shortlist and want to compare it.

What should you prepare before the first call?

Before the first call, write down the one or two marketing jobs you most need done, the channels that bring work today, your monthly budget including ad spend, and the name of the person who will review the work each week. Providers give sharper answers to a clear brief, and you can compare them on the same terms.

Half a page is enough. Include your last three months of enquiries if you have them, even as a rough count, and a list of the accounts you already own: website, Google Business Profile, ad accounts, email platform, social pages. Send the same brief to every provider on the shortlist.

Agree early on what value looks like, because value is one of the most common reasons clients leave an agency. Setup's 2025 Marketing Relationship Survey found 61% of clients who ended an agency relationship cited dissatisfaction with value, and 44% said the agency did not understand the business2, though from only about 100 responses, mostly larger US brands.

Which questions test the system itself?

Four questions test whether a provider runs real AI systems: ask for a live run of one, ask what it reads and produces, ask where a person reviews it, and ask how the work splits between systems and people. A provider with working systems answers these in minutes, with a screen share of a real run.

1. Can you show me the system running on a real business?

Ask for a live or recorded run, start to finish: the input it read, the draft it produced, the review, and what shipped. A screenshot of a tool's dashboard is a picture of software. A run shows a working process.

2. What does each agent read, and what does it produce each month?

Good answers are specific: "the search agent reads Search Console, your service pages and the questions customers ask, and produces two updated pages and a tested question set each month." Vague answers ("our AI optimises your marketing") mean the provider has not built it yet.

3. Where does a person review the work, and who is that person?

Someone should approve everything published under your name, every first message to a prospect and every change to ad spend. Ask for a name and a review day. Review matters because unchecked AI output goes wrong: KPMG and the University of Melbourne's 2025 New Zealand snapshot, surveyed over summer 2024-25, found 51% of NZ employees had relied on AI output at work without evaluating its accuracy and 46% had made mistakes because of AI3. The arrangement is called human in the loop.

4. Which tasks do your systems do, and which do people do?

The split should match the price. If systems do most of the production but the retainer is billed in hours at the usual agency rate, the saving stays with the provider. That can still be a fair deal, as long as you know it.

Which questions settle ownership and data?

Four questions settle ownership and data: who owns the agents and accounts, how customer data is handled under the Privacy Act, which AI tools see your data and where they store it, and how outbound lists and messages meet spam law. Written answers here protect you more than any contract clause.

5. Who owns the agents, accounts, prompts and data?

Your website, ad accounts, analytics, email platform and social pages should be in your name from day one, with the provider added as a user. Ask whether the agents, prompts and workflows built for you are yours to keep. If they stay with the provider, you are renting the system, which should be reflected in the price.

6. How do you handle our customer data under the Privacy Act?

In New Zealand, the Privacy Act 20204 applies to any business handling personal information. The Privacy Commissioner's information privacy principles include storage and security (IPP 5) and disclosure outside New Zealand (IPP 12)5, and the Commissioner expects a privacy impact assessment before AI tools are used, updated regularly6. Ask whether the provider has done one for the tools it will use on your data.

In Australia, the OAIC says the Privacy Act generally applies to businesses with annual turnover above A$3 million7, plus some smaller ones such as health providers. The OAIC's 2024 guidance recommends that organisations do not enter personal information into publicly available generative AI tools8, and from 10 December 2026, businesses covered by the Act that use personal information in automated decisions that could affect people's rights or interests must describe them in their privacy policies9. The guide to privacy law for AI marketing in NZ and Australia covers both countries in more depth.

7. Which AI tools and models touch our data, and where is it stored?

Ask for a list: the AI models, the automation platform, the CRM, and the country each one stores data in. Your customers care about this too. The Privacy Commissioner's 2026 survey of 1,245 New Zealanders found 82% want more say in how their information is collected and used10.

8. How do outbound lists and messages meet spam law?

If the provider runs outbound, ask where the lists come from. IPP3A, in force since 1 May 2026, requires reasonable steps to tell people when their information was collected indirectly11, which covers enriched lead lists. Under New Zealand's Unsolicited Electronic Messages Act, Consumer Protection says commercial messages need consent, accurate sender details and a free unsubscribe that works within five business days, with fines up to $500,000 for businesses12. Australia's Spam Act sets similar rules, enforced by the ACMA.

Which questions test results and reporting?

Two questions test results: which numbers the provider will report and how often, and what will exist at the end of the first month that does not exist today. The answers should name enquiries, their sources and cost per enquiry, plus concrete outputs such as pages, emails or a working agent.

9. Which of our numbers will you report, and how often?

Two to four numbers, monthly, starting with enquiries and cost per enquiry. For AI search work, add a fixed set of customer questions tested in ChatGPT, Perplexity and Google before and after. Reports built on impressions and follower counts alone describe activity.

10. What will exist at the end of month one?

Good answers are lists: a positioning file, five updated service pages, a review request flow, a weekly report. If the answer is "a strategy" with nothing running, ask when the first system goes live.

Which questions cover price, term and exit?

Two questions cover the commercial side: what the price is in writing and what sits outside it, and what happens when you leave. A provider that answers both before the second meeting is usually easy to work with. One that avoids them often makes leaving hard too.

11. What is the price, in writing, and what costs sit outside it?

Get a written figure with GST, ad spend, software subscriptions and setup fees listed separately. Few providers publish prices: Digital Nomads HQ's June 2026 audit of 140 Australian agencies found 78% publish no pricing and only 8% a genuine rate card13. The AI marketing cost guide for Australia and New Zealand gives market ranges to compare against.

12. If we stop, what do we keep and how does the handover work?

Ask for the minimum term, the notice period and a written handover list: documentation, prompts and workflows, logins transferred, data exported. A provider confident in its work makes leaving simple.

What are the red flags?

The clearest red flags are a provider that cannot show a working system, sets up accounts in its own name, guarantees rankings or AI mentions, has no named reviewer, reports impressions only, or avoids written answers on price, data and exit. Two or more on one shortlist entry is usually enough to drop it.

  • A capability deck in place of a demonstration
  • Your ad account, website or analytics owned by the provider
  • A guaranteed first-page ranking or a guaranteed mention in ChatGPT
  • Nobody named to review work before it reaches customers
  • A twelve-month lock-in with no handover clause
  • Vague answers about which AI tools see customer data
  • Outbound lists of unknown origin, or no unsubscribe in sample emails
  • Claims about results without a client you can call

Claims matter legally too. The Commerce Commission lists Fair Trading Act fines of up to $600,000 per offence for a business14, and marketing an AI provider writes for you goes out under your name.

How do you score providers side by side?

Score each provider from 0 to 3 on six criteria, multiply by the weight, and add up. The table puts the most weight on the running system, ownership and review, because those are hardest to fix after you sign. Price matters, but compare it only between providers that pass the first three.

CriterionWeightScores 3 whenScores 0 when
Running system (questions 1, 2)3Showed a real run with inputs, outputs and reviewSlides only
Ownership (5, 12)3Everything in your name, written handover listAccounts or systems stay with the provider
Human review (3, 4)3Named reviewer, set review day, clear system and people splitNo one named
Data and privacy (6, 7, 8)2Tools listed, storage stated, privacy assessment done, list sources explainedUnclear which tools see data
Measurement (9, 10)2Enquiries and cost per enquiry monthly, month-one outputs listedImpressions only
Price and terms (11)1Written price, extras listed, short minimum termNo written price

The maximum is 42. A provider under 25 needs a good reason to stay on the list. Ask for references as well: Setup's 2025 survey found 82% of clients who found an agency through word of mouth were satisfied, against 70% through internet search2. For named providers in each country, explore the overviews of AI marketing providers in New Zealand and AI marketing providers in Australia.

Is the best choice different for a small business?

For a small business, the best choice is usually the provider that does the fewest things well, on your accounts, with a short first term. A 5 to 50 person business rarely needs every service line at once. One or two lines that run every week, with a person reviewing, beat a broad retainer that stalls.

Agencies with media buying teams remain the right answer for large paid media budgets and brand campaigns. For a business whose main gap is being found in search and AI answers, keeping content going and following up on enquiries, a focused AI marketing agency for small business, or a build your own team runs, usually fits better. If you are comparing options in New Zealand, the page on AI marketing services in New Zealand sets out what is available locally; Australian readers can start from AI marketing services for Australian businesses.

Where Flow AI fits

Flow AI provides AI marketing services for businesses in New Zealand and Australia. Amber Lan, an AI marketing engineer in Auckland, builds the AI agents that run search, content, outbound and reporting with your team, then hands them over. Flow AI welcomes the same twelve questions as any provider on your list.

Its systems are public in the skill library and the free agent templates, so you can answer most of questions 1 and 2 before a call. Each build starts with a written scope and is priced per project. It ends with a documented handover, and from there your team owns it and keeps it running. Companies hiring a marketing engineer or GTM engineer can reach Amber through the hiring page.

The short version

Choose an AI marketing provider by what it will show you: a working system, a named reviewer, accounts and data in your name, privacy answers in writing, a report built on enquiries, and a clear exit. Score each shortlist entry on the table above and drop anyone under 25.

Keep the first engagement small and short, on one or two service lines, and expand once the numbers move.

To put Flow AI through the twelve questions, talk to Amber.

Questions owners ask

What is the best AI marketing agency for a small business?

The best AI marketing agency for a small business is one that shows its system running on a real business, names the person who reviews your work, keeps every account and file in your name, reports enquiries and cost per enquiry, and puts its price and exit terms in writing. Size and awards matter less than what it will show you before you sign.

What questions should I ask an AI marketing agency?

Ask it to show the system running, explain what it reads and produces, name who reviews the work, say who owns the agents, accounts and data, explain its data handling under privacy law, list the numbers it reports, state the price in writing, and describe the handover if you leave. Twelve questions cover it, and a good provider answers them in writing.

How can I tell if an AI marketing provider is real?

Ask for a live or recorded run of its system on a real business, showing the input, the output and the review step. Gartner estimated in 2025 that only about 130 of thousands of vendors selling agentic AI are real, so labels are unreliable. A provider with working systems can show one in minutes; a provider with a label tends to show slides.

Does the Privacy Act apply to AI marketing in New Zealand?

Yes. The Privacy Act 2020 applies to any New Zealand business handling personal information, including customer lists, enquiries and lead data used by AI tools. The Privacy Commissioner expects a privacy impact assessment before AI tools are used, and IPP3A, in force since 1 May 2026, requires telling people when their information was collected indirectly, which covers enriched lead lists.

What are the red flags when hiring an AI marketing agency?

Red flags include no demonstration of a working system, accounts or ad accounts set up in the agency's name, guaranteed rankings or guaranteed AI mentions, reports built on impressions alone, no named reviewer, long lock-in terms without a handover clause, and vague answers about which AI tools see your customer data. Two or more of these usually justify walking away.

Should a small business choose an AI marketing agency or a big agency?

A small business with several channels to run and little in-house time often gets more from a focused AI marketing provider. A traditional agency with a media team remains the better choice for large paid media budgets and brand campaigns. Decide by the work you need done this year, then use the same twelve questions on both.

Sources

Every figure links to its primary source. Checked 13 October 2026.

  1. Gartner estimated in June 2025 that only about 130 of the thousands of vendors selling agentic AI are real, in a warning about "agent washing", and predicted over 40% of agentic AI projects will be cancelled by the end of 2027 gartner.com
  2. Setup's 2025 Marketing Relationship Survey found 61% of clients who ended an agency relationship cited dissatisfaction with value, and 44% said the agency did not understand the business setup.us
  3. KPMG and the University of Melbourne's 2025 New Zealand snapshot, surveyed over summer 2024-25, found 51% of NZ employees had relied on AI output at work without evaluating its accuracy and 46% had made mistakes because of AI assets.kpmg.com
  4. Privacy Act 2020 legislation.govt.nz
  5. The Privacy Commissioner's information privacy principles include storage and security (IPP 5) and disclosure outside New Zealand (IPP 12) privacy.org.nz
  6. the Commissioner expects a privacy impact assessment before AI tools are used, updated regularly privacy.org.nz
  7. the OAIC says the Privacy Act generally applies to businesses with annual turnover above A$3 million oaic.gov.au
  8. The OAIC's 2024 guidance recommends that organisations do not enter personal information into publicly available generative AI tools oaic.gov.au
  9. from 10 December 2026, businesses covered by the Act that use personal information in automated decisions that could affect people's rights or interests must describe them in their privacy policies oaic.gov.au
  10. The Privacy Commissioner's 2026 survey of 1,245 New Zealanders found 82% want more say in how their information is collected and used privacy.org.nz
  11. IPP3A, in force since 1 May 2026, requires reasonable steps to tell people when their information was collected indirectly privacy.org.nz
  12. Consumer Protection says commercial messages need consent, accurate sender details and a free unsubscribe that works within five business days, with fines up to $500,000 for businesses consumerprotection.govt.nz
  13. Digital Nomads HQ's June 2026 audit of 140 Australian agencies found 78% publish no pricing and only 8% a genuine rate card digitalnomadshq.com.au
  14. The Commerce Commission lists Fair Trading Act fines of up to $600,000 per offence for a business comcom.govt.nz