August 31, 2026 · Clemens

Ads don't help you get qualified customer feedback

We built Context Goblin, an AI code review platform, and got our first customer validation through internal company usage, Hacker News, and Product Hunt. Then we launched ad campaigns to get broader traction for the product. Here's what we learned from running Google Ads and Reddit Ads for a developer tool in the age of AI.

TLDR: Ads guarantee a good number of impressions and clicks, but they don't help you get qualified customer feedback. From our experience, running ads helps you understand edge cases, B2C customer flow, and landing page performance. The most important factors seem to be the landing page and the onboarding flow.

Ads don't help you get qualified customer feedback

We got 1,700 and 2,200 conversions for Google Ads and Reddit Ads respectively, on a budget of €200. Those numbers translated into around 80 signups from people of every background - we saw some genuinely interested personas, but mostly new gmail accounts without any intention to try the product.

Compared with the roughly 50 people we reached through friends, colleagues, and coworkers - who actually tested the product in the environment it was designed for and gave detailed, valuable feedback about their individual needs - this broad customer acquisition looks weak.

Interestingly, the highest engagement for the ad campaigns came from India and Bangladesh, which matches the experience of startup friends of ours (govideo, talecut).

Tweak your ad presence before launching a campaign

The biggest mistake we made before launching the ad campaigns was not tightening our ad presence to the absolute limit. Although we thought we already had solid product design and marketing, our landing page and ad banners converted very poorly.

The landing page has to be completely clear about what the product does and how to use it. We lost many first-time visitors because of the GitHub connection our product requires.

Build with AI, but don't market with AI

Being at the forefront of AI development, we leverage coding and non-coding agents wherever we can. But developers, as a community, are quickly growing sick of AI-written text, and they punish even slightly vibed texts with downvotes. The lesson: we'll keep building and iterating with coding agents, but we'll stop completely letting AI write text meant for humans to read.