Zeroclick is an answer engine optimization platform I built for local businesses and the agencies that serve them. The premise is that local search stopped ending in ten links and started ending in one AI generated answer that names three or four businesses. If you are not in that answer you do not lose the click, you never existed. Zeroclick measures whether AI names your business, tracks that over time, and tells you what to fix.

The core of the system is a prompt simulation engine. It generates the questions real customers actually ask at local granularity, meaning this service in this city in this zip rather than a national average, then runs them on a weekly or daily schedule against ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews. Every returned answer is parsed for brand mentions, competitor mentions, cited source domains, and sentiment. That rolls up into a visibility score, a per engine breakdown, and a share of presence ranking against local competitors.

The hard parts were not the interface. LLM answers are non deterministic and unstructured, so mention detection had to handle name variants, partial matches, and businesses with generic names without generating false positives. Scores had to stay stable week over week so a client sees real movement instead of model noise. And running tens of thousands of prompts a month across five engines meant a scheduled job pipeline with rate limit handling, retries, cost controls, and result caching.

Sitting on top of the measurement layer is the action layer, which is a fix checklist ranked by impact, a site audit, a content library, and one click publishing to WordPress. The whole thing is multi tenant for agencies, with brands, markets, and prompt quotas metered per plan and add ons purchasable without a plan change.