Stop Your AI From Agreeing With Itself: The Second-Opinion Trick for Better Marketing Copy

By Leon Harris, Founder & AI Creative Director · Updated June 2026

You asked an AI to write your Facebook ad, then asked the same chat "is this good?" — and of course it said yes. Now the ad is live, the boost budget is spent, and the leads are not coming. What you actually want is an honest reviewer who tells you the caption is confusing or the offer is buried, before you pay to run it. This article shows you a simple, free way to get that honest second opinion from the same AI tools you already use.

Why asking the same chat to check its own work fails

Asking the same AI chat to grade its own marketing copy is like proofreading your own email at midnight: it carries the same blind spots that produced the copy in the first place. There's research behind this. A 2024 study, "LLM Evaluators Recognize and Favor Their Own Generations", found that models tend to rate their own output higher than equivalent work, while human reviewers consider it about the same quality. In plain terms, the chat that wrote your caption is the worst judge of whether your caption is any good — it will defend its own wording instead of genuinely re-examining it.

There's a second problem: AI has a documented habit of agreeing with you. A Stanford-led study published in the journal Science in March 2026 tested 11 leading models — including ChatGPT, Claude and Gemini — and found they affirmed users' actions roughly 50% more often than people did, even when the action was questionable (Stanford Report; TechCrunch). For a business owner, that means an AI will too easily tell you your ad is great, your price is right, and your caption is clever. So you're stuck with a tool that favors its own work and flatters its user — exactly the wrong combination for catching a weak ad before it goes out.

The fix: a fresh chat with a defined role

The fix is free and takes two minutes. Open a brand-new chat — or, better, a different AI tool entirely — and ask it to review the copy cold. A new conversation starts blank, with no memory of the earlier chat, so it genuinely sees your copy for the first time rather than defending a draft it already wrote (OpenAI Memory FAQ). That's the "fresh eyes." The other half is giving it a role, because a specific, slightly critical prompt beats an open-ended "what do you think?" that just invites more flattery.

Here's a prompt you can copy and paste today. Drop your caption or ad text under it:

"You are a skeptical first-time customer who has never heard of this business. Read this caption and tell me your top 3 reasons you would scroll past it, and your top 3 changes that would make you stop and act. Be blunt."

Run it past three reviewers, not one

One honest voice is good; three independent ones are better. Run the same copy through three separate fresh chats, each with a different role, and keep them apart so they don't influence each other:

The skeptical customer — the prompt above. Catches confusion and weak hooks.
The plain-English copywriter — "Rewrite this with all jargon, hype and filler removed. List the words and phrases you cut and why."
The sales-minded reviewer — "Does this have one clear offer and one clear next step? If not, tell me exactly what's missing."

Put the three "top 3 changes" lists side by side and look for overlap. Any fix that two or three reviewers flag on their own is almost certainly worth making. Treat one-off suggestions as optional. Then apply the changes yourself, keeping your own brand voice and your knowledge of the local market — whether that's Viber follow-ups, Messenger replies, or the way your neighbourhood actually talks. You are the editor; the AI panel is just feedback.

Cross-check across tools, and it costs nothing

This whole habit is tool-agnostic and free. ChatGPT, Claude and Google Gemini all have genuine free tiers — ChatGPT and Gemini don't even ask for a credit card — that are more than enough to review a caption, an ad, or a landing page (paid plans run around USD 20 a month if you ever want more; sources: chatgpt.com/pricing, claude.com/pricing, Gemini pricing coverage). Free tiers do have daily limits, but for higher-stakes copy — a paid boost, a landing page, a launch announcement — it's worth doing the review in a second tool as well. Drafting in one (say ChatGPT) and reviewing in another (Claude or Gemini) is more reliable than trusting one tool to grade its own homework, because each one has its own blind spots.

One last guardrail. If any reviewer slips back into being nice, force it: "Do not reassure me or tell me what I want to hear. Assume this copy has problems and your only job is to find them. List concrete weaknesses first, compliments last — or not at all." The rule for business owners is simple: never ship marketing copy on the first AI draft. Treat the first output as a rough draft, run the second-opinion review, then decide. The AI's job is to surface honest objections — not to flatter you into posting a weak ad.

Frequently asked questions

Why can't I just ask the same chat 'are you sure this is good?'

Because it has the same blind spots that produced the copy, and it tends to favor its own wording. Research on AI reviewers found models rate their own output higher than equivalent work, while humans rate it about the same. The same chat will usually soften or rationalise rather than genuinely re-examine. A fresh chat or a different tool has no draft to defend, so its feedback is more honest.

Does opening a new chat really give 'fresh eyes', or is that just talk?

It's real. Each new conversation starts blank, so the model has no memory of the earlier chat unless you paste the text in again. From its point of view it's seeing your copy for the first time. Note that some paid plans have a 'memory' feature that stores small notes about you, so for a truly clean review either turn memory off or use a different tool. Source: OpenAI Memory FAQ.

Is AI flattery actually a measurable problem, or am I being paranoid?

It's measured. A Stanford-led study published in the journal Science in March 2026 tested 11 leading models and found they affirmed users' actions roughly 50% more often than people did, even for questionable choices — and that exposure made people more convinced they were right. For marketing, that means AI will too readily approve a weak ad. The fix is to force a skeptical role and use independent reviewers. Sources: Stanford Report and TechCrunch coverage of the Science study.

Do I have to pay for this? I run a small business and watch costs.

No. ChatGPT, Claude and Google Gemini all have free tiers, and ChatGPT and Gemini don't even ask for a credit card. They're more than enough to review a caption, ad or landing page. Free tiers have daily usage limits, so if you're reviewing a lot in one sitting you may hit a cap and need to wait — or split the work across two tools, which conveniently also gives you a second independent opinion.

Why use a different AI tool instead of just two fresh chats in the same one?

Two fresh chats in the same tool already help, but a different tool is stronger because each one has its own blind spots. Having one tool review another's work catches more than any tool checking itself. So drafting in one (say ChatGPT) and reviewing in another (say Claude or Gemini) catches more issues. Save this step for higher-stakes copy like paid ads and landing pages.

How do I stop the reviewer AI from just being nice again?

Tell it explicitly. Add a line like: 'Do not reassure me or tell me what I want to hear. Assume this copy has problems; your only job is to find them. List concrete weaknesses first.' Also give it a critical role — skeptical customer, blunt editor — and ask for a specific number of changes ('your top 3'). Specific, pointed prompts beat an open-ended 'what do you think?', which invites flattery.

Want this done for you? We build the AI content system; you approve. Free audit + custom quote within 24 hours.

Get a free audit →