ChatGPT Says Your Google Ads Campaign Is "Great" (Even When It's Losing Money)
ChatGPT is insanely useful for marketers. It's fast, confident, and it can turn a messy export into neat recommendations in seconds. The problem? It will also tell you a campaign is "performing exceptionally well" based on ROAS alone, even if it's losing money net.
The Core Issue
If you blindly follow ChatGPT's scaling advice based on ROAS alone, you'll scale yourself into a cashflow problem. This article explains why that happens, what ChatGPT is missing, and the exact checks you need to run before increasing budget.
The Real Problem: ChatGPT Optimises for "Plausible", Not "Profitable"
When you paste a Google Ads report into ChatGPT and ask "Is this campaign good?", you're usually giving it performance metrics like ROAS, CPA, conversions, conversion value, clicks, and cost.
So ChatGPT does what it's designed to do: it gives you a confident, helpful answer based on the inputs.
If ChatGPT sees:
- 5x ROAS
- Strong conversion volume
- Stable CPA
It will often recommend:
- → Increasing budget
- → Scaling bids
- → Expanding keywords
- → Loosening targets
- → Pushing Performance Max harder
Because in marketing theory land, that's logical. In business reality land, that can be a disaster.
Why ChatGPT Tells You What You Want to Hear
ChatGPT isn't trying to mislead you. It's doing exactly what it was trained to do.
It mirrors your framing
If you ask "Is 5x ROAS good?", you're framing the decision as ROAS = success. So it replies accordingly. It won't automatically say "That depends on margin, returns, fulfilment cost, and cash timing" unless you force it to.
It's trained to be helpful, not commercially accountable
Humans get punished for bad decisions: cash runs out, stock goes sideways, finance escalates, founders panic, teams churn.
ChatGPT doesn't get punished by your P&L. So it will confidently recommend scaling a campaign that is quietly losing money net.
It doesn't understand your unit economics unless you provide them
Most Google Ads exports don't include COGS, shipping cost, fulfilment cost, returns rate, payment fees, discounts, contribution margin, or cashflow timing.
So the model literally cannot know profitability. It's guessing based on incomplete information.
The Classic Trap: 5x ROAS That Still Loses Money Net
Let's make it real. A campaign shows 5x ROAS, £10,000 spend, £50,000 revenue attributed. ChatGPT says: "This is great, scale."
But your actual unit economics look like this:
55% gross margin?
Maybe
30% margin?
More common
20% margin?
Very common in DTC
Now add: shipping subsidies, fulfilment and pick/pack, returns and exchanges, payment fees, discounting, customer service overhead, warehousing costs, stock write-offs.
Suddenly your "5x ROAS" can be:
ROAS is a ratio. It's not a profit metric.
ROAS Is Not Profit: Here's What It Ignores
1. Contribution margin
If your contribution margin is 20% and your ROAS is 5x:
£1 ad spend returns £5 revenue
20% margin on £5 = £1 gross profit
You just broke even before returns, shipping subsidies, and overhead
So the campaign can look "great" while doing nothing for the business.
2. Returns (especially fashion, beauty, premium goods)
Returns don't show up cleanly in Google Ads reporting. A campaign can "print ROAS" and still lose money after returns, exchanges, failed deliveries, and cancellations.
3. Shipping and fulfilment costs
Many brands subsidise shipping to convert. That cost often isn't in ad platform reporting, so ROAS can be inflated versus reality.
4. Discounts
ROAS looks amazing when discounting is doing the conversion work. If you're paying for traffic that only converts at 20% off, you're not scaling marketing. You're scaling margin erosion.
5. Cash timing and payment terms
Even if a campaign is "profitable", it can be cashflow-toxic if inventory is paid upfront, customers pay later (BNPL), returns lag, or chargebacks hit later.
Scaling can create a cash crunch even when ROAS looks strong.
The Bigger Problem: Using ChatGPT for "Strategy"
This is where it gets worse. Many teams are now doing: "Give me growth ideas" → pick the most exciting ones → ship them → wonder why nothing improves.
LLM ideas are usually: generic, unpriced, unprioritised, and consequence-free.
They sound smart until you overlay the thing most businesses forget exists: context.
Context is: margin structure, stock constraints, team bandwidth, compliance, creative capacity, category dynamics, channel saturation, customer behaviour.
Once you add those constraints, half the "ideas" go from clever to commercially unhinged.
How to Use ChatGPT Properly for Google Ads (Without It Costing You Money)
ChatGPT is useful if you treat it like a junior analyst, not a decision-maker.
Step 1: Ask better questions
"Is this campaign good?"
"Based on these metrics AND margin, returns, and fulfilment costs, is this campaign profitable? If we scale budget 20%, what breaks first?"
Step 2: Give it the missing commercial inputs
Feed it:
- • Gross margin or contribution margin %
- • Average order value
- • Returns rate
- • Shipping cost per order
- • Fulfilment cost per order
- • Discount rate
- • Payment fees %
- • Target profit per order (or POAS target)
Then it can actually reason properly.
Step 3: Use POAS or profit per order as the decision metric
ROAS answers:
"Did ads generate revenue?"
POAS answers:
"Did ads generate profit?"
If your goal is scaling profitably, ROAS is not enough.
Step 4: Pressure-test scale like a CFO
Before scaling budget, check:
- • Does profit scale with spend?
- • Does CAC rise with volume?
- • Do we have stock depth?
- • Will returns spike with broader traffic?
- • Can ops handle higher order volume?
- • Will cash timing break?
Scaling should be a controlled increase, not a wrestling entrance.
A Simple "Should We Scale?" Checklist
If a campaign has strong ROAS, you still need to confirm these before increasing budget:
Profitability
- □ What's contribution margin by SKU/category?
- □ What's profit per order after returns?
- □ What's blended CAC vs margin?
Quality
- □ Are we driving new customers or just harvesting brand demand?
- □ Are we discount-dependent?
- □ Are we cannibalising organic/email?
Operational Constraints
- □ Stock depth and lead times
- □ Fulfilment capacity
- □ Customer service load
- □ Returns processing ability
Cashflow Constraints
- □ Inventory payment terms
- □ Payout delays
- □ Return windows
- □ Chargeback risk
If you can't answer those, don't scale. You're just making the numbers louder.
Final Thought: ChatGPT Isn't Wrong, Your Question Is
ChatGPT is not the enemy. But if you ask it a ROAS-only question, you'll get a ROAS-only answer.
That's how you end up scaling a campaign that looks great on a dashboard and feels awful in the bank account.
Scale profit. Not screenshots.
Related Reading
Want a commercial review of your Google Ads account?
We audit accounts based on profit contribution, not platform metrics. If you're spending £10k+ per month and want to know what's actually working, let's talk.
Request a Commercial Audit