ChatGPT Ads and GEO: The Protocol to Test Your Paid vs Earned Share
Summarize this article with AI
The Protocol Starts with a Call
A client calls me on a Tuesday morning. He invested 8 000 $ in ChatGPT ads over 30 days. Result: 214 clicks, 3 direct conversions. He asks me a simple question: « Would I have gotten these clicks for free? »
Silence.
I pause before answering. What’s at stake here is the arbitrage between paying and earning. Cost per click means little to me.
OpenAI and Go Fish Digital just published a cross-analysis on Search Engine Journal. Their angle: ChatGPT Ads on one side, the earned organic visibility on the other. And in the middle, a gap in measurement.
I see this gap in 9 out of 10 clients. They launch AI ads. They look at the dashboard. They count clicks. But they lack a baseline for their earned visibility. It’s impossible to know if the ad adds or replaces.
Additions.
Replacements.
Two different realities. One is profitable. The other is a leak.
Heather Campbell, from Search Engine Journal, puts her finger on what many sensed was coming: now, we must measure and test the arbitrage. That’s the missing link for growth teams.
I’ve been building this protocol for 6 months. With my clients, I observe orders of magnitude: 47 AI queries tracked, 8 000 $ in testing, 14 days to decide. Concrete.
Before paying, map what you already earn. Otherwise, you’re buying air.
And air is expensive.
In the coming weeks, the most advanced teams will know how to read the difference between an earned click and a bought click. The budget alone will cease to be decisive.
This protocol costs little. It demands rigor.
Three numbers to remember. 47 queries. 14 days. 2 groups. That’s it.
I’ll show you the phases, the metrics, the pitfalls. At the end, you will know how to launch your own test.
In 2026, those who win will know what the ad brings on top. This skill is built with tests. Opinions alone fall short.
What ChatGPT Ads Really Changes
ChatGPT Ads is more than a simple ad insert. According to the Search Engine Journal article, OpenAI has opened a format where sponsored suggestions appear in ChatGPT responses. They are labeled. They are contextual. They directly compete with the organic citations the model chooses on its own.
The landscape is changing.
Yesterday, a single question: « Does ChatGPT talk about me? »
Today, two questions: « Does ChatGPT talk about me? » and « Am I paying for it? »
Earned visibility, called earned AI visibility in the article, depends on evidence the model considers solid. Go Fish Digital talks about a GEO evidence audit. Concretely: your product page, your reviews, your structured data, your external mentions.
Paid visibility, for its part, comes from your budget. It’s a direct lever. But it coexists with the other.
The problem: teams treat these two levers separately. Mistake.
I often compare this to two engines on the same plane. If you push one engine while ignoring the thrust of the other, the plane drifts.
The protocol I propose measures both at the same time. On the same queries. With the same goal: spotting what adds incremental value.
Incremental.
Not total.
A paid click that replaces an organic click carries a different price.
Another change: data. With ChatGPT Ads, OpenAI is starting to provide click data. It’s new. Still partial. But sufficient for a clean test.
In 6 months, this data will be richer. Start now. Organic positions are built right now.
Every ChatGPT response where your brand is absent is an opportunity a competitor pays for.
Every response where you appear organically is an arbitrage asset.
This protocol turns that asset into a number.
The article’s source specifies: ad formats in ChatGPT are still young. Performance data varies by sector. Build your own benchmarks instead of trusting external ones.
The 4-Phase Protocol
Phase 1. Mapping AI queries.
I select 47 strategic queries. Not 500. Not 5. 47. Why 47? Because that’s the number I tracked with a client on a recent deployment. 47 are enough to decide. Beyond that, it’s noise.
Each query maps to a purchase or decision intent. Example: « best payroll software for small businesses ». You note the current position in ChatGPT. Paid? Organic? Absent?
It’s a snapshot. Nothing more.
Phase 2. Earned visibility baseline.
For 7 days, I capture ChatGPT responses on these 47 queries. In an ad-free state. I note citations and sources. I take time-stamped screenshots.
This baseline is your zero point. It makes comparison possible.
A tip: set up a shared file. One line per query. Column « organic citation », column « cited competitor », column « source ». In 7 days, you have a snapshot.
An operational detail: for capture, switch ChatGPT sessions, enable browsing mode if you have it, and note the time. Responses vary. Reproducibility matters.
Phase 3. Launch ChatGPT Ads on half the queries.
I keep the other half as a control group. That’s the basis of a clean test. You pay on 23 queries. The other 24 remain in pure organic.
The choice of the 23 is deliberate. I take high-volume queries, those where the purchase decision happens. The 24 control queries are similar. Apples to apples.
Phase 4. Measuring incrementality.
After 14 days, I compare the two groups. Clicks, conversions, organic citations. If the paid group delivers equivalent results to the control, the ad is a replacement, not an addition.
The DOSE framework, taught by Guillaume Attias at BMO Academy, sheds light on this protocol: dopamine comes from rapid testing. Measurable feedback every 7 days drives progress. Action is immediate.
You test, you measure, you decide. Fourteen days, and you know.
The Metrics That Really Matter
Which metrics to track? Avoid impressions. They inflate dashboards and drain budgets.
I have 5. Ranked by priority.
| Metric | What it measures | Frequency |
|---|---|---|
| AI share of voice | Citations of your brand across 47 queries | Weekly |
| ChatGPT Ads click-through rate | Paid clicks / sponsored impressions | Daily |
| Assisted conversions | Visits initiated by AI then converting elsewhere | Over 7 days |
| Incremental cost per conversion | Difference between paid and control groups | End of test |
| GEO evidence | Sources ChatGPT uses to cite | Monthly |
The details.
1. Share of voice in AI responses. Across your 47 queries, how many times is your brand cited? That’s your organic baseline. One client went from 3 citations out of 47 to 11 citations in 90 days, organically. Just by strengthening GEO evidence. Observed order of magnitude.
2. Click-through rate from ChatGPT. If OpenAI provides this data, track it. Otherwise, use tracking URLs, promo codes, or dedicated landing pages.
3. Assisted conversions. An AI click can later trigger a Google search. Track the correlation over 7 days in your analytics tools. Direct conversion alone is insufficient.
4. Incremental cost per conversion. It’s the difference between the paid group’s cost and the control group’s conversion. Not the total cost. This is the number that justifies or rejects the budget.
5. GEO evidence. Go Fish Digital recommends auditing the sources ChatGPT uses: reviews, product pages, external articles. They feed earned visibility and keep the organic engine running.
A bad metric can be more costly than the absence of metrics.
I often see teams looking at average CTR. 2,3 %. They applaud. Except those 2,3 % include clicks that would have come anyway. What matters: how many extra.
To measure that, a control group is required. Non-negotiable.
The impressions trap: they explode, morale rises, but arbitrage remains invisible. I prefer 100 incremental clicks to 10 000 impressions.
The test gives its dopamine in the comparison with the control, not in the raw number.
The Arbitrage Pitfalls
Pitfall #1: believing organic is free.
It is far from free. Producing content, structuring data, getting reviews. All of that costs time and money.
The question that matters: what is the marginal cost of each channel?
Sometimes, ads are cheaper than the editorial effort needed to appear organically.
Sometimes, it’s the opposite.
Only the test tells you.
Pitfall #2: ignoring cannibalization.
A paid click that replaces an organic click is burned money. Not a gain. I saw a client go from 12 % organic share of voice to 9 % after launching an aggressive ChatGPT Ads campaign. Sponsored suggestions pushed organic citations lower. Cost per conversion doubled.
Observed order of magnitude with him, not a universal law.
Cannibalization is measurable, not a theory. Simple example: you launch an ad on query A. Your organic share of voice on A goes from 8 citations out of 20 to 5 citations out of 20. Sponsored suggestions took the place. You lost 3 citations rather than gaining 3.
Pitfall #3: testing in the absence of a control group.
In the absence of a control, you compare a period with ads to an ad-free period. But seasons, events, and model updates distort everything.
The control isolates the variable.
Pitfall #4: stopping at the first result.
A 14-day test gives a trend. Not a definitive truth. AI models evolve fast. ChatGPT Ads is recent. Formats change. Your protocol must run continuously.
Every 6 weeks, I relaunch a 14-day burst on 15 pivot queries. That’s the routine.
The protocol’s dopamine is not the final result. It’s the short loop: test, measure, adjust.
Pitfall #5: confusing direct conversion with contribution.
A visitor may click an AI ad, read, leave, and return 5 days later via Google. If you track only direct conversion, you underestimate the impact.
Conversely, a paid click may cannibalize a conversion that would have come organically. You overestimate the impact.
The protocol decides.
The data decides.
The teams I see often confuse activity with progress. The protocol turns that activity into measurable returns.
Launch Your Test in 14 Days
Here is the exact plan. 14 days. Not a promise. A sequence.
Day 1 to 2. Choose 15 AI queries. Transactional keywords, comparison questions, competitor brand queries. Note their current position.
Day 3 to 5. Capture the organic baseline. Open ChatGPT, type the queries, take screenshots. Note competitor citations.
Day 6 to 7. Set up ChatGPT Ads on 7 or 8 queries. The other half remains the control. Prepare unique tracking URLs.
Day 8 to 14. Measure every day. Clicks, conversions, citations. Look for gaps between the paid and control groups.
At the end, three possible decisions.
Either the ad adds incremental value. You double down on winning queries.
Or the ad replaces organic. You reduce the budget and invest in GEO evidence.
Or there is no difference. You wait. The protocol will have at least avoided a blind spend.
That’s arbitrage.
Not an intuition.
A test.
The tools: ChatGPT access, a ChatGPT Ads editor, a spreadsheet, tracked URLs. Nothing more.
The minimum budget? Order of magnitude: 2 000 $ over 14 days. Enough for 7 queries in a competitive market. In a local niche, 500 $ may be enough.
The mistake: spending more before having a baseline.
Baseline.
Control.
Gap.
Three pillars.
After the test, document everything. Results, screenshots, gaps. That’s your reference for deciding. It sharpens with each iteration.
And you, how many AI queries do you already earn unknowingly?
AI Visibility Audit in 47 Points
I look at your citations in ChatGPT, I trace your strategic queries, I give you the complete protocol. Commitment-free. Let’s talk.
Book a strategic call — 45 minFrequently Asked Questions
Is ChatGPT Ads available in France?
Rollout in waves. Test with an OpenAI account. Check the criteria. And prepare the measurement protocol upstream.
How much does a minimal test cost?
Count 2 000 $ to 5 000 $ over 14 days, depending on competition. My advice: start small. Data is worth more than budget.
Can I measure earned without advertising?
Yes. Each week, you record your citations on 20 queries. You copy ChatGPT responses. That’s your baseline.
Which metrics should I avoid?
Advertising impressions, that’s all. They reveal nothing about arbitrage. Look at clicks, assisted conversions, share of voice.
Does GEO replace classic SEO?
No. Google still carries weight. GEO adds on top. The protocol compares three channels: classic search, earned AI, paid AI.
