Meta Is Hallucinating Your Ad Copy (And Lying About It)

What Founders Are Getting Wrong About Meta's "Off" Switches

Founders are reporting a pattern that should terrify anyone running paid acquisition: Meta is serving AI-generated ad copy they never wrote, with prices they never approved, despite every automation toggle being switched off.

Here's the specific failure mode: A D2C brand running ads for a $69 product started getting customer complaints about pricing mismatches. Turns out Meta's system had been serving ads claiming the product cost $29—a price that never existed in their copy, their catalog, or anywhere in their ad account. The founder had entered one copy variation. One. Advantage+ creative was disabled at both the ad and account level. Meta fabricated the copy anyway and burned budget on it.

This isn't a bug. It's a feature you didn't consent to.

The reality most founders miss: Meta's automation doesn't respect your settings the way you think it does. The platform is optimizing for its own objectives—time on platform, engagement, revenue per impression—not your unit economics or brand integrity. When founders say "all automation is off," they're often wrong. They just don't know which levers actually control the system anymore.

The Real Problem: Platform Incentives Misaligned With Founder Outcomes

Let's be direct about what's happening here. Meta operates the largest auction-based ad platform in the world. Their revenue model depends on maximizing CPMs and keeping advertisers spending. AI-generated creative variations increase engagement rates, which increases CPMs, which increases Meta's revenue. The fact that some of those variations include hallucinated pricing or fabricated claims? That's an externality Meta doesn't pay for—you do, in refunds, customer service overhead, and brand damage.

The incentive structure is clear: Meta wants you to give their AI maximum creative freedom because it performs better for Meta. Sometimes it performs better for you too. But when it doesn't, you're the one holding the bag.

This creates a trust problem. If you can't trust that "off" means off, you can't rely on platform settings as your control mechanism. You need a different verification architecture.

What Actually Controls Meta's Creative Freedom 🔒

Here's what founders need to understand about Meta's creative systems in 2025:

Account-level Advantage+ settings don't govern everything. These toggles control whether Meta suggests automation to you, not whether background systems can modify your creative. Think of them as preference settings, not hard constraints.

Ad-level creative settings have more teeth, but still leak. Disabling creative optimization at the ad level should prevent copy variations, but as we're seeing, the system doesn't always respect these boundaries—especially when Meta's AI detects what it thinks is a "performance opportunity."

Catalog data can override manual ad copy. If you're running catalog ads or have a catalog connected to your pixel, Meta may pull pricing from there even when you've manually specified different copy. This is supposed to prevent mismatches, but it creates them when catalog data is stale or incorrectly structured.

Preview environments lie. The ad preview you see in Ads Manager is not necessarily what gets served. Meta generates variations in real-time based on user signals, placement, and device type. You're approving a template, not a final asset.

The only reliable controls are: manual review of actually-served ads (not previews), catalog hygiene, and aggressive alert systems for pricing mismatches.

How to Actually Lock Down Your Creative

1. Build a monitoring system that doesn't trust Meta

Stop relying on Ads Manager as your source of truth. You need independent verification of what's actually being served to users.

Set up a weekly audit calendar:

• Screenshot random served ads from the Facebook Ad Library (search your Page name, filter to active ads)
• Compare Ad Library versions to what's in Ads Manager
• Run test purchases from different devices/locations to see what landing experience users actually get
• Cross-reference any promotional pricing in ads against your actual offer stack

This takes 30 minutes a week and catches discrepancies before they become customer service nightmares.

2. Sanitize your catalog data

If Meta has access to a catalog with incorrect pricing, it will use it. This is especially dangerous if you:

• Run seasonal promotions without updating catalog pricing
• Have test products with placeholder prices
• Import feeds from a PIM system that includes internal cost data

Action: Audit your product catalog monthly. Every SKU should have current, customer-facing pricing. If you run promotions, handle them via overlay copy or custom labels—don't change base catalog pricing unless it's a permanent change.

3. Use creative lockdown settings correctly

Here's the specific toggle sequence that actually works (as of Q1 2025):

Campaign level:

• Keep Advantage campaign budget OFF (use ad set budgets)
• This prevents Meta from reallocating budget to AI-modified variants without your knowledge

Ad set level:

• Keep Advantage+ audience OFF if you're testing specific segments
• Dynamic creative should be OFF unless you're explicitly testing it with throwaway budget

Ad level:

• Multiple text options: Set to 1 variation only
• Disable "Optimize text per person" (this is the key one)
• Use manual placements or carefully curated automatic placements—avoid "Advantage+ placements" for high-stakes offers

4. Implement a price verification prompt system

If you're using AI to generate ad copy (and you should be, just not Meta's), you need quality control. Here's a copy-paste-ready prompt you can use with Claude or ChatGPT to generate ad variations that won't hallucinate pricing:

You are writing ad copy for [product name]. Follow these constraints exactly:

PRICING RULES:
- The ONLY approved price is $[your actual price]
- Never invent discounts, sales, or alternative pricing
- If mentioning price, use the exact format: "$[price]"
- Do not use phrases like "starting at" or "as low as" unless I explicitly provide a range

VERIFICATION STEP:
After generating copy, list every price mentioned and confirm it matches $[your actual price]. If you generated any other price, flag it as an error.

Generate [number] ad variations for [campaign objective] targeting [audience]. Focus on [key benefit/angle].

This prompt structure forces the AI to self-verify pricing before you review it, catching hallucinations at the generation stage.

The Broader Pattern: Automation Without Accountability

This Meta incident is part of a larger pattern in performance marketing. As platforms automate more of the creative and bidding process, they're shifting responsibility for outcomes onto founders while retaining control of the actual mechanics. You're accountable for ROAS, but you don't control what gets served or to whom.

The answer isn't to avoid automation—manual campaign management doesn't scale and often performs worse. The answer is to build verification and oversight systems that treat platform automation as untrusted by default.

That means:

• Never assuming a platform setting does what it says without verification
• Building independent monitoring that doesn't rely on platform reporting
• Treating AI-generated anything (copy, creative, targeting) as drafts requiring human review
• Structuring campaigns so that failures are contained and detectable

The founders who win in this environment aren't the ones who trust platform tools blindly. They're the ones who verify everything and engineer their campaigns to fail visibly when automation goes wrong.

Action Checklist

Use this to audit your Meta setup this week:

• [ ] Check Facebook Ad Library for your Page—screenshot 10 random active ads
• [ ] Compare Ad Library versions to Ads Manager versions—note any discrepancies
• [ ] Review product catalog pricing for every SKU—flag any mismatches with current offers
• [ ] Verify Advantage+ creative is OFF at ad level for all campaigns with specific pricing claims
• [ ] Set calendar reminder for weekly ad spot-checks (15 min)
• [ ] Implement price verification prompt for any AI-generated copy workflow
• [ ] Create a shared doc for customer complaints mentioning pricing—review weekly for patterns
• [ ] Test purchase flow from a cold device—verify landing page pricing matches ad claims

Get Your Free Growth Audit

Running paid acquisition for a subscription app, Shopify brand, or home service business and worried about what's actually being served to your prospects?

Advanced App Marketing offers free 30-minute growth audits where we'll review your Meta account setup, identify automation risks, and show you exactly what's leaking budget or damaging trust.

We're a lean AI growth agency that works exclusively with founders who need performance, not reports. Book your audit at advancedappmarketing.com/audit—we'll tell you what's broken and how to fix it.

---