ChatGPT ad targeting is garbage

I person been experimenting pinch buying ChatGPT ads for my event seating duty software. Here are the wide results from 05-Aug-2026 to 28-Aug-2026:

ChatGPT recommends bidding location astir by £2 per click, but I bid betwixt £0.15 and £0.07 and was still capable to get traffic. However 0.46% is simply a shockingly debased conversion rate. Given what I cognize astir the life customer worth and instal to acquisition rates for PerfectTablePlan, I’m not adjacent to making immoderate money connected these ads.

Comparing engagement pinch different sources successful the aforesaid period:

So the problem is intelligibly the ChatGPT advertisement traffic, alternatively than my website/product. It seems 1 of respective things mightiness beryllium happening:

1. OpenAI is charging for much clicks than they are sending.
2. My discourse hint is bad.
3. My ads are bad.
4. A precocious proportionality of the clicks are from bots, not humans.
5. The postulation is very severely targetted.

The clicks reported by OpenAI are adjacent to what Google Analytics reports, truthful we tin norm 1 out.

My discourse hint seems uncontroversial, truthful that rules retired 2.

The ads are very akin to ones I person been moving successful Google Adwords for years, truthful we tin norm 3 out.

I reached retired to integer advertisement fraud master Dr Augustine Fou, portion measurement done the experiment, to inquire him whether he thought bots mightiness beryllium responsible for a precocious proportionality of the clicks. He was benignant capable to thief maine instrumentality my website pinch fouanalytics.com. Looking astatine clip connected page, rodent movements and different metrics were capable to found that a mostly of the clicks were apt humans. Here you tin spot the Fouanalytics analysis, pinch 30% showing ‘many reddish flags’ aliases ‘some reddish flags’ for being bots.

So location mightiness beryllium immoderate bot traffic, but not capable to relationship for the unspeakable conversion rate. So we tin norm 4 out.

That only leaves 5. The postulation is poorly targetted. This is supported by information from Fouanalytics. I wrangled the information successful Easy Data Transform to get a histogram of the clip connected the landing page, wherever each bin is 1 second:

Fouanalytics besides shows that 88% of the postulation moved their mouse, but only 10% clicked anything. Which is simply a further denotation of poorly targetted traffic.

Targeting is everything erstwhile it comes to advertising. Clearly ChatGPT knows really to nonstop maine targetted traffic, because their free referrals person an bid of magnitude amended than their paid ads. Maybe they are truthful hopeless to capable the yawning void successful their finances that they are generating arsenic galore paid clicks arsenic they can, without caring excessively overmuch astir the quality?

I would beryllium fascinated to cognize successful what discourse they are showing my ads. But location is nary measurement for maine to find that out. Consequently, I don’t cognize really I tin amended this conversion rate. Even if I did, I wouldn’t beryllium hopeful astir improving the conversion complaint by the bid of magnitude required for profitability.

Here is Dr Fou’s conclusion:

“Best practices confirmed by this research — ever usage existent conversions arsenic the measurement of success. The problem pinch ChatGPT ads was NOT bots; it was debased worth visitors (ones that arrived connected the landing page but did thing other and left). It’s not wholly clear WHY these visitors are truthful debased successful worth (e.g. were the ChatGPT ads simply targeted very poorly?) Invest your advertisement budgets successful sources that not only output much clicks to your site, but astir importantly much existent conversions.” Dr Augustine Fou