How do we use AI to manage Ads campaigns? Three workflows explained

One question comes up in almost every scoping meeting: "Are you actually using AI on our campaigns?" The answer is yes, but rarely where you might imagine. At Growth Room, AI doesn't choose the strategy or the budget. It helps us manage Ads campaigns faster and with greater rigor, at three specific moments: before launch, during ad production, and daily during account monitoring.

Here are our three AI workflows, step by step, with what each one really changes and what remains in the hands of a consultant.

Key takeaways

• We use AI at three stages: intent mapping, creative production and testing, and account monitoring.
• In every workflow, the AI prepares and proposes, while a consultant validates and decides. No budget is moved without human approval.
• Nothing works without reliable data: tracking, HubSpot, and GA4 are connected before the first euro is spent.
• The method these workflows accelerate is proven: at Poiscaille, it cut the acquisition cost by 7 and generated over 2,000 subscriptions.
• It applies as-is to ChatGPT Ads, available in France since August 24, 2026, where you target conversations rather than keywords.

Why use AI to manage Ads campaigns in 2026?

Because ad platforms have already automated part of the job, and the value has shifted elsewhere. Google Ads' automated bidding adjusts the price of every click much better than a human could do manually. We detail how they work in our guide on Google Ads bidding strategies.

But an ad platform's AI only sees its own ecosystem. It ignores your margins, what happens to your leads in the CRM, and how your prospects phrase their needs. That is precisely where we bring in our own workflows.

What does the ad platforms' AI already do, and what is left to be done?

In short, we let the ad platforms do what they do best, and we focus our use of AI on what they will never do for us.

Workflow 1: How does AI help us map out intent before launch?

A campaign is won before it ever goes live. If you target the wrong queries, no amount of optimization will fix it. Our first workflow is therefore used to build a map of intent : who is searching for what, with which words, and at what stage of their decision-making process.

It takes place in four stages:

1. We gather the raw material: search terms from existing accounts, questions heard by sales teams, customer feedback, and competitor pages.
2. The AI groups these phrases by intent and decision stage: discovery, comparison, and purchase.
3. It suggests what is missing: long-tail phrases, questions asked in natural language, and exclusions to plan for.
4. The consultant makes the final call. Which intents deserve an ad group, which should be added as negative keywords, and which are off-target.

This logic didn't start with AI. At Poiscaille, for example, generic queries like "online fishmonger" fueled the engagement phase, while localized queries like "fish delivery Paris" were reserved for conversion. With AI, the reasoning remains the same, but we carry it out across a much larger volume of phrases without spending days on it.

How does this workflow change for ChatGPT Ads?

On ChatGPT Ads, there are no more keywords. You describe to OpenAI Ads Manager the types of conversations where your offer belongs: these are the context hints. The system then decides on its own where to serve the ads.

Our workflow translates almost directly. We map out the questions prospects ask ChatGPT, step by step, just as we do for our target prompts in GEO. Then, we test several families of hypotheses in parallel. On this platform, we estimate that the right context accounts for 80% of performance.

Workflow 2: How do we produce and test creatives with AI?

A good ad is rarely found on the first try. You have to try several, and that is where AI saves the most time. Our second workflow organizes creative testing in five steps:

1. The brief. The consultant defines the angle, the promise, the evidence supporting our claims, and the brand tone.
2. Generation. The AI adapts each angle into multiple headlines and descriptions, formatted exactly for each ad platform.
3. Curation. The consultant removes anything that rings false, rewrites what needs to be, and selects a few variations per hypothesis.
4. Testing. The variations are launched in parallel, each tied to a clear hypothesis.
5. Review. We keep what wins on CPA or ROAS, cut the rest, and then start a new cycle.

Step 3 is non-negotiable. An AI can write a catchy headline that is completely inaccurate. No ad goes live without human review.

Why is AI useful for short-form content?

Because the constraints are extreme. On ChatGPT Ads, an ad must fit into a headline of about 16 characters and a description of about 32 characters. Finding the right formula in such a small space requires a lot of trial and error, and AI produces these iterations in minutes.

What doesn't AI replace in the creative process?

The perspective on imagery and branding. In the case of Poiscaille, our most useful recommendations concerned visuals and landing pages:

• adding the logo and a call to action to all ads;
• creating a carousel to compare subscription plans;
• testing step-by-step recipe videos and customer testimonials;
• simplifying the subscription page menu to keep only the essentials.

This type of recommendation comes from analyzing the ads with the best CPA, not from a text generator.

Workflow 3: How does AI monitor accounts and prepare for budget adjustments?

An Ads account can drift silently. A rising CPA, a broken tag, or ad fatigue: the problem is often spotted too late, after the budget has already been spent. Our third workflow is based on anomaly detection and a weekly summary.

1. Data from ad platforms, GA4, and HubSpot are reconciled, from the initial click to the qualified lead.
2. The AI compares metrics against their usual levels and flags any deviations.
3. It generates a summary: what has changed, potential explanations, and areas to investigate further.
4. The consultant verifies, diagnoses, and decides: cut, correct, or reallocate.

Which signals do we monitor?

• A cost per acquisition that deviates from the target.
• Abnormal spending on a campaign or ad group.
• A sudden drop in conversions, which most often indicates a tracking issue.
• A discrepancy between the leads reported by the ad platform and the qualified leads in the CRM.
• A declining click-through rate, a sign of ad fatigue.

On what criteria do we decide to shift budget?

Based on the cost per qualified lead and the revenue generated, never on impressions. The AI summary prepares thebudgetary decision, but it does not make it.

The reason is simple: raw numbers can be misleading. A campaign that performs poorly on a last-click basis might be the one fueling your pipeline the most, as we explain in our article on the true profitability of B2B Ads. Resolving cases like this requires understanding the client's sales cycle. That is the consultant's job.

What results does this method produce? The Poiscaille example

An AI workflow is only as good as the method it accelerates. Ours boils down to three verbs: test often, measure everything, reallocate quickly. The Poiscaille case is the most striking illustration, with a customer acquisition cost divided by 7.

Poiscaille sells seafood subscriptions sourced directly from French fishermen. We structured their acquisition strategy on Google Ads and Meta into three phases: awareness, engagement, and conversion. Between November 21, 2022, and January 31, 2023, the campaigns generated over 2,000 subscriptions, with the cost per subscription decreasing month over month.

‍

These figures only count a single subscription per customer, even though subscribers stay for an average of 8 to 10 months. The actual profitability is therefore even higher.

What drove down costs for Poiscaille?

Three actions, repeated throughout the mission:

• A/B testing across numerous audiences, both cold and warm: interests, lookalike audiences, and remarketing;
• regular monitoring of the cost per subscription, broken down by platform and campaign;
• concrete recommendations for ads and landing pages.

These are precisely the three actions our workflows now automate. The method remains the same, but the pace has increased: more hypotheses are tested, and deviations are spotted sooner.

The full details of the strategy, audiences, and winning ads can be found in the complete Poiscaille case study.

What does the AI decide, and what does the consultant decide?

The AI does not decide anything. It prepares, proposes, and alerts. Every decision that impacts your budget or your brand image is made by a consultant. The table below summarizes the division of roles.

What are the limitations of AI in managing Ads?

We see three, and we prefer to state them upfront.

• It doesn't know your business. Your margins, sales cycle, and brand constraints must be provided to it, otherwise, it operates blindly.
• It can be confidently wrong. A plausible hook is not necessarily a true one, which is why systematic review is required.
• It depends on data quality. Without tracking reliable, it analyzes noise.

This last point explains our high standards for measurement. Our COO was a team leader at Fifty Five, and every consultant is trained to install and interpret tracking for their accounts. AI workflows come after, never before.

What happens to these workflows on an advertising platform as new as ChatGPT Ads?

They are even more valuable there. On Google Ads or Meta, years of history guide decisions. On ChatGPT Ads, which opened in France on August 24, 2026, everything remains to be learned: which conversations to target, which phrases work within 16 characters, and what cost per qualified lead to expect. Mapping intentions, running multiple tests, and monitoring metrics closely: this is what allows for rapid learning without wasting budget.

Our first campaigns are currently running, with the OpenAI pixel and Conversions API connected to HubSpot from day one. If you are wondering what role this platform can play alongside your current campaigns, our ChatGPT Ads agency will scope out the test with you, run the pilot, and provide you with a data-backed verdict.

‍