Our agency uses ChatGPT and other AI tools to write and run our ads. What should we ask it, and should we pay less?
TofuAnswer by the Tofu team · checked
Ask four things: which AI tools touch your account, what of your data goes into them, who checks the output before it runs, and who answers for mistakes. Using AI is normal. The risk is your customer lists or CRM exports going into a personal chatbot plan that can use them for training. Pay less only if you pay for hours and AI cut them, or the agency now does less. On a flat fee, ask for more tests or faster turnaround instead, and keep judging the agency on qualified pipeline.
Ask Tofu next
Pick a question and Tofu Ads Agent answers it in chat.
- What should an AI-use clause in our paid media agency contract say?
- Which Google Ads assets in my account were written by Google's AI?
- Which changes did someone other than our team make in our ad accounts last month?
- How do I check whether our agency's ad copy makes claims we can't back up?
- Should we move from an hourly agency fee to a flat fee now that agencies use AI?
As of November 2026.
Platform AI vs chat assistants vs AI connected to your accounts
Agencies use AI in three different ways, and each needs different questions.
| The ad platforms' own AI | Chat assistants (ChatGPT, Claude, Gemini) | AI connected to your ad accounts | |
|---|---|---|---|
| What it does | Writes headlines, descriptions and images inside the platform, such as Google's generated and suggested assets | Drafts copy and briefs, suggests keywords, summarizes reports and exports | Reads and changes campaigns through an API, an MCP server or an agent |
| What of your data it sees | What's already in your ad account | Whatever the agency pastes or uploads | Whatever the connection's access allows, often the whole account |
| Main risk | Ads that say things nobody approved | Your data on a plan that can use it for training or human review | Bid, budget or targeting changes nobody approved |
| What to ask | Which automated asset settings are on, and who reviews new assets | Which tool, which plan, whether training is off, what data goes in | What it can change without a person, and where changes are logged |
| Effect on the agency's hours | Small: the platform does the writing | Can cut drafting and reporting time | Can cut routine checks and edits |
When the ad platforms' own AI is fine
It's fine when someone reviews what it writes. Google tells advertisers: "You should review any generated or suggested assets to ensure that they're accurate, not misleading, and not in violation of any Google advertising policies or applicable laws before publishing them." It adds that generated assets "aren't guaranteed to be approved."
Google's suggested assets are "available to all advertisers by default." If you've opted into asset optimization, Google says "These suggested assets don't need to be reviewed before saving." So ask your agency whether that setting is on and who looks at new assets. B2B claims go wrong easily: a security certification you don't hold, a price you don't offer, an integration you don't have.
When chat assistants are fine
They're fine for first drafts, ideas and summaries that hold no personal or confidential data. They're not fine for customer lists, CRM exports or call notes unless the agency uses a business plan that doesn't train on them. The plan matters more than the brand:
- Claude: Since Anthropic's August 28, 2025 update, people on Claude Free, Pro and Max choose whether their chats are used to train models, and Anthropic keeps data for five years for those who allow it. The update doesn't apply to Claude for Work (Team and Enterprise) or the API.
- Gemini: In the Gemini app, Google says it uses your activity to improve its services, "including training generative AI models," when Keep Activity is on. It also says "A subset of chats are reviewed by human reviewers" and asks people not to enter confidential information they wouldn't want a reviewer to see. For Google Workspace, Google says "Workspace does not use customer data for training models without customer's prior permission or instruction."
- OpenAI: OpenAI's developer docs say data sent to its API "is not used to train or improve OpenAI models" unless you opt in, and that abuse monitoring logs, which can include prompts and responses, are kept for up to 30 days by default. OpenAI's pages on ChatGPT's personal and business plans couldn't be read for this page, so ask your agency which ChatGPT plan it uses and how its data controls are set.
When AI connected to your ad accounts is fine
It's fine when it can only read, or when every change to spend or targeting waits for a person to approve it. Ask what access the connection has, whether it can launch or raise budgets on its own, and where you can see a log of what it changed. Our page on connecting Claude or ChatGPT to ad accounts explains how these connections work. For which tasks are safe to hand to software and which should stay with a person, see what to automate and what to keep human in B2B paid media.
What should we ask our agency?
- Which AI tools do you use on our account, and for which tasks?
- Which plan or license is each one, and is training on our data turned off?
- What of our data goes into them? Customer lists, CRM exports, revenue figures, call recordings?
- Who reviews AI-written ads, images and landing page copy before they go live?
- Which automated settings in our ad platforms are on, such as Google's automated asset optimization?
- Can any AI tool change bids, budgets or targeting without a person approving it?
- Who pays if an AI mistake wastes budget or gets an ad or account flagged?
- What has AI saved you on our account, and where did that time go?
If a tool is sold to you as an "AI agent," the test in AI ads agent vs ad automation software helps you check what it really does.
What should the contract say?
- A list of approved AI tools. Adding a new one needs your OK.
- What data may go in. No customer lists, CRM exports or confidential plans in any tool that can train on them or let people review them.
- Your sign-off rules. A person reviews every ad before it runs, and spend or targeting changes above a set amount need your approval.
- Who carries the cost of errors. For example, a credit for spend wasted by an unreviewed change.
- Your ownership. Ad accounts, assets and outputs belong to you. Our page on judging a B2B PPC agency covers account ownership and the numbers to ask for.
If the EU's GDPR covers your customer data, this isn't optional. When your agency handles personal data for you, it is your "processor." Article 28 of the GDPR says a processor "shall not engage another processor without prior specific or general written authorisation of the controller." Under a general authorisation, the agency must tell you about any processor it adds or replaces, so you can object. An AI vendor that processes your customer list for the agency is one.
Should we pay less?
It depends on what your fee pays for:
- Hours. If you pay by the hour or for a set number of hours, the savings should show up in fewer hours or more work done in them. Ask for both on the invoice.
- A percent of ad spend. The fee follows your budget, not the agency's effort, so AI doesn't change it. If the fee already feels too high, look at the model itself: see flat fee vs percent of ad spend.
- A flat retainer for a scope. If the scope still gets done well, the price is fair. Ask for more within it (more tests, faster changes, better reports), or renegotiate at renewal if the work clearly shrank.
- Results. If you pay for qualified leads or pipeline, how the agency gets there is its business.
You aren't alone in asking. The ANA (Association of National Advertisers) said in November 2025 that "clients expect to see a reduction in agency fees as a result of more AI adoption."
Don't cut the fee for strategy, offer testing and judgment. Those were never the hours AI saves. Judge the agency on qualified leads and pipeline, as in how to judge a B2B PPC agency, not on how the ads were made.
Sources (8)
- How to build a Performance Max asset group using generative AI (Google Ads Help), checked November 9, 2026
- About suggested assets (Google Ads Help), checked November 9, 2026
- Updates to Consumer Terms and Privacy Policy (Anthropic, August 28, 2025), checked November 9, 2026
- Gemini Apps Privacy Hub (Gemini Apps Help, updated October 26, 2026), checked November 9, 2026
- Generative AI in Google Workspace Privacy Hub (Google Workspace Help), checked November 9, 2026
- Data controls in the OpenAI platform (OpenAI developer docs), checked November 9, 2026
- Regulation (EU) 2016/679 (GDPR), Article 28 (Official Journal L 119, Publications Office of the EU), checked November 9, 2026
- How AI Is Impacting Agency Operations, Compensation, and More (ANA, November 20, 2025), checked November 9, 2026