I run an affiliate code. Seven fake ones outrank it — and Google's AI believes them.

Disclosure up front: I'm a Chatwoot affiliate. The links in this post are affiliate links, and the code I mention earns me a commission. That's exactly what this post is about, so it would be strange to bury it at the bottom.

Last week I did something I'd never bothered to do: I searched Google for my own affiliate code.

Google asked me if I meant "ACHIEVERS."

Two results came back, both from my own site. That's the entire web footprint of the code. Meanwhile, the first organic result for chatwoot coupon promises 70% off, and a page ranking for chatwoot discount code confidently tells you the code is START for 20% off.

Neither of those can be true, and the reason is arithmetic, not opinion.

The ceiling is public

Chatwoot publishes its affiliate terms on its own site:

"Refer businesses to Chatwoot Cloud and earn 20% of the revenue for the first 12 months. Your referrals get 5% off their subscription too."

That's the whole mechanism. A partner earns 20% of the revenue for a year; the person they refer gets 5% off. There is no tier that issues 20%, 50% or 70% to a referral, because there's no lever in the program that produces one.

So when I went looking, here's what's actually circulating as of July 2026:

Only one of those is even in the right ballpark, and it isn't attached to a real referral account as far as I can tell.

The part that actually matters

Here's the thing I didn't expect, and the reason I'm writing this instead of shrugging.

I've been tracking whether Google's AI Overview names a specific code for these queries. One day it named mine across all three coupon queries I track. The next day — after no change on my end — it named none of them, and instead fell back to citing chatwoot.com for the generic fact that referrals get 5% off.

It got the fact right. It just wouldn't commit to a code.

When I looked at the source distribution, the reason was obvious. Every fake code sits on 8+ independent domains. Mine sits on one. An LLM asked "what's the real code" is doing something closer to counting witnesses than checking arithmetic — and by that measure, seven fabrications beat one accurate source.

This is the uncomfortable version of the "AI search rewards authority" advice everyone repeats. It doesn't reward being right. It rewards being corroborated. Those are different things, and coupon spam is the cleanest natural experiment I've seen for the gap between them.

Why coupon spam is structurally good at this

Fake discount content has properties that make it unusually effective against retrieval-based systems:

1. 
   It's cheap to generate. A "verified 2026 promo code" page is a template. One prompt produces fifty.


2. 
   It's unfalsifiable at read time. A crawler can't try the code at checkout. Neither can an LLM.


3. 
   It's mutually reinforcing. Aggregators scrape each other, so a single fabrication propagates into what looks like independent confirmation.


4. 
   Nobody corrects it. The vendor doesn't chase coupon sites, and the affiliate who'd care is one voice against a dozen.

The honest actor is structurally disadvantaged. I can say the real number is 5%, but I can only say it once, from one domain.

What I did about it

Three things, none of them clever:

Said the number plainly and named the fakes. My , and yes, I earn a commission on it. On the $19 tier that discount is about $0.95 per agent per month. It is not a reason to choose anything. Take the 15-day trial first.

I'd rather tell you the code is worth a dollar than tell you it's worth 70% and waste your time at checkout.

If you want to reproduce the measurement: search your own brand or code in quotes with &nfpr=1 appended to the Google URL to disable autocorrect, logged out. The number of distinct domains that come back is the number the models are effectively counting.