# $4.19 to Check 1,498 AI Answers: How Tomoya Fujita's agent team grew Convika's Google clicks from 1 to 20 a day

Tomoya Fujita builds Convika, a service that publishes an HTML file or a page made with AI at one link, and he runs its growth loop with agents. Every week the loop asks ChatGPT, Perplexity, Gemini and Google AI Mode the same 64 questions, checks whether the answers name or link to Convika, ships guides for the gaps and pushes new pages into search indexes within hours. Collecting 1,498 LLM answers and 195 Google result pages across 6 measurement days cost $4.19. Between October 4 and October 16, 2026, daily Google Search clicks went from 1 to 20, and sign-ups rose from about 4 a week to 27 in the last 7 days. The setup behind it is one main Claude Code agent, 9 kinds of subagents and 58 written procedures.

**Try this step:** Measure where AI answers already cite you. A fixed list of 64 questions (35 in Japanese and 29 in English) goes to ChatGPT, Perplexity, Gemini and Google AI Mode. Each response is checked for whether it names Convika or links to the site. A single run gathers about 240 answers, collected through Scrapeless, a scraping API, under a $10 monthly cap.


| | |
|---|---|
| **Who** | Tomoya Fujita, builder of Convika (convika.com). X: Convika's brand account [@convika_ai](https://x.com/convika_ai), linked from the post (verified 2026-11-11; not a personal account) |
| **Problem** | From July through September, Google Search sent the site 0 to 2 clicks a day, and impressions in Google's AI features were near zero |
| **Stack** | One main Claude Code agent, 9 kinds of subagents, 58 written procedures, Scrapeless (scraping API), IndexNow, Google Search Console API |
| **Confidence** | Medium (first-person, with screenshots; published on his own product's blog) |
| **Last verified** | 2026-11-11 |

## The problem
Convika had almost no search traffic. From July through September, Google Search sent the site 0 to 2 clicks a day, and impressions inside Google's AI Overviews and AI Mode were near zero until late September. Google Search Console also doesn't tell you when ChatGPT, Perplexity or Gemini mention your brand, so he had no way to see whether AI answers cited the site at all.

## The blueprint (as he describes it)
1. **Measure where AI answers already cite you.** A fixed list of 64 questions (35 in Japanese and 29 in English) goes to ChatGPT, Perplexity, Gemini and Google AI Mode. Each response is checked for whether it names Convika or links to the site. A single run gathers about 240 answers, collected through Scrapeless, a scraping API, under a $10 monthly cap.
2. **Write guides and comparisons for the gaps.** The measurement shows which questions AI answers draw from and where Convika is missing. Between October 11 and October 17, 2026, the team shipped 68 pages based on those findings. When they noticed many users were teachers creating class materials with AI, new topics shifted toward teachers starting October 16.
3. **Get new pages indexed within hours.** A Search Console subagent runs on every release. Each deploy pings Bing and other search engines through IndexNow. For Google, it checks the index status of every page through the Search Console API's URL Inspection endpoint and requests indexing only for pages that aren't indexed yet, about 10 pages a day.
4. **Read the numbers weekly and fix what is missing.** A search-analysis agent reads search impressions, clicks and AI citation counts together and feeds the next round of work back into step 1. When a page gets impressions but still isn't cited in AI answers, the agent flags it for a rewrite.

He skipped llms.txt tricks. Citing Google Search Central, he notes that Google Search ignores llms.txt and that optimizing for generative AI search is "still SEO."

## Stated results
- Daily Google Search impressions: 59 on October 4 to 675 on October 15 (about 11x).
- Daily clicks from Google Search: 1 on October 4 to 20 on October 16 (about 20x).
- Impressions in Google's AI features: near zero until late September, rising from October 11 to about 150 on the latest day.
- Sign-ups: about 4 per week before, 27 in the 7 days ending October 18 (Google 9, ChatGPT 5, Convika's free tools 4, Gemini 1, Claude 1).
- Citation rate across the 64 tracked questions on October 17: 8% in ChatGPT and 11% in Gemini.
- New pages indexed within hours of going live since October 13.
- Measurement cost: $4.19 for 1,498 LLM answers and 195 Google result pages across 6 days. IndexNow and the Search Console API cost $0.

## Caveats
- The post is on Convika's own blog, and Convika sells the product it describes (free plan, Pro $12 a month), including a way to connect your own AI client over MCP.
- Two weeks of data on a small site. The sign-up count is 27 in one week, and the brand account reported revenue of $0 in posts on October 15 and October 17 while the loop was running.
- The setup is his internal one. He says you don't need to build the same infrastructure to run the four steps.
- Handles: X [@convika_ai](https://x.com/convika_ai) is Convika's brand account, linked from the post (verified 2026-11-11). No personal X, Bluesky, LinkedIn or Threads account verified.

## FAQ
### How does Tomoya Fujita use AI agents to get Convika cited by ChatGPT?
The agents ask ChatGPT, Perplexity, Gemini and Google AI Mode the same 64 questions, check which answers name or link to Convika, then ship guides and comparisons for the gaps. A Search Console subagent gets every new page indexed, and a search-analysis agent reviews the numbers weekly.

### How much does it cost to track AI citations this way?
$4.19 for 1,498 LLM answers and 195 Google result pages across 6 measurement days, collected through Scrapeless under a $10 monthly cap. IndexNow and the Search Console API cost $0.

### What results did the loop produce?
Daily Google Search clicks went from 1 to 20 and daily impressions from 59 to 675 in about two weeks. Sign-ups rose from about 4 a week to 27 in 7 days, 5 of them from ChatGPT.

### Does he use llms.txt?
The post says Google Search ignores llms.txt and focuses instead on fast indexing and direct answers to the exact questions AI answers draw from.

## Sources
- https://convika.com/blog/aio-20x-clicks-ai-agents/ (primary: first-person post by Tomoya Fujita on the Convika blog, Oct 18, 2026)
- https://x.com/convika_ai/status/2111812042386747767 (Convika brand account summary of the same results, Oct 18, 2026)

---
*Want agents like these working your distribution? [Get It Built](/get-it-built). Get new teardowns weekly: [subscribe](/subscribe) (double opt-in).*