Building AI visibility for YepIts.ai — what we shipped, what worked, what's still in progress

The starting point

YepIts.ai is a YouTube video summarizer built in Zagreb, Croatia. The product works — paste a YouTube URL, get key takeaways + clickable timestamps in ~10 seconds. Free tier is 3 summaries/day; Pro is $7/mo.

But when we ran a baseline citation audit on August 21, 2026 — asking ChatGPT, Perplexity, Claude, and Gemini "what's the best YouTube video summarizer?" — YepIts.ai was mentioned in zero out of one test response. ChatGPT's response named Eightify, NotebookLM, Glasp, NoteGPT, Recall, and YouTube's native AI. YepIts.ai was not in the picture.

This is the GEO problem GEO exists to solve: a working product, invisible to the recommendation systems users are now asking.

What we found when we looked under the hood

Before adding anything, we audited the existing YepIts.ai setup. Two big findings shaped the work:

1. The technical GEO foundation was already strong

• robots.txt explicitly allowed all major AI crawlers (GPTBot, ChatGPT-User, Claude-Web, PerplexityBot, Google-Extended, etc.) — 10+ user-agents with explicit Allow rules
• llms.txt deployed in proper spec format with product pages, blog, FAQ, documentation sections
• sitemap.xml listing all main pages and blog posts
• Schema markup: three JSON-LD blocks (Organization, SoftwareApplication, FAQPage) in the homepage
• Six existing comparison blog posts (vs Eightify, vs Notta, vs NoteGPT, plus a "Best YouTube Summarizer 2026" roundup)

2. The technical foundation was not being served correctly

This is where it got interesting. The Dockerfile was missing COPY public/ ./public/, which meant the rich AI-aware robots.txt never made it to the production container. What users (and AI crawlers) were getting was the minimal fallback from the React build directory — User-agent: * / Allow: / with no crawler-specific permissions.

Second issue: robots.txt was being served with the wrong MIME type and the entity-disambiguation sameAs block in the Organization schema included 4 broken URLs (github.com/yepitsai, crunchbase, linkedin, producthunt — all 404'd) and one Twitter handle that belonged to a different person entirely.

Third and biggest issue: the blog was broken in production. The YepIts.ai repo has two parallel src/ directories — one at the repo root with the real code, and one under frontend/src/ with a stub. Vite builds from frontend/, which meant the "real" Blog component and all 6 comparison posts were never bundled. Anyone visiting yepits.ai/blog/yepits-vs-eightify got a "coming soon" placeholder.

What we shipped (August 2026)

• Week 1 — Audit + bug fixes
  Ran citation baseline test (YepIts.ai invisible in ChatGPT). Found 3 bugs: missing Dockerfile copy step, broken sameAs links, stub blog component shadowing the real implementation. Fixed all 3.
• Week 2 — Entity presence
  Created LinkedIn company page, new @yepits_ai X account (the @yepitsai handle was taken), and a Wikidata entity with 8 statements linking the GitHub repo, X, and LinkedIn. All wired into the Organization sameAs schema block.
• Week 3 — Content
  Wrote 3 new comparison posts: vs Glasp, vs Recall, vs YouTube's built-in AI. Targeted at the exact competitors ChatGPT named in the baseline. Updated sitemap.xml so AI crawlers see the new URLs immediately.

Final state (live on yepits.ai as of August 21, 2026)

• 9 blog posts live and accessible, including the original 6 and 3 new comparisons
• 4 entity profiles linked in Organization sameAs: X (@yepits_ai), GitHub (AntoskaJede/yepitsai), LinkedIn (yepitsai), Wikidata (Q141163371)
• Schema markup: Organization (with 4 sameAs), SoftwareApplication (with offers), FAQPage (7 Q&As)
• robots.txt allows 10+ AI crawlers explicitly
• llms.txt with full product description in llmstxt.org spec
• sitemap.xml with 10 URLs

What we expect to see (and when)

Per the GEO research (Princeton GEO paper, Aggarwal et al. 2024) and our own testing on other sites, AI model citation growth follows this rough timeline:

• Perplexity: 2-6 weeks from content going live — it browses the web in real-time
• ChatGPT: 2-4 months from content going live — its training/index updates on a slower cycle
• Claude: 2-4 months — similar to ChatGPT, with entity-heavy inputs helping
• Google AI Overviews: 4-8 weeks — depends on crawl frequency

So the citation growth from this August work will show up roughly:

• September 2026: Initial movement in Perplexity and Google AI Overviews (assuming crawl catches up)
• October-November 2026: ChatGPT and Claude picks up as training data refreshes
• December 2026: Mature baseline measurement — what does YepIts.ai look like in AI recommendations after 4 months?

What we'll measure

We'll re-run the same citation audit on October 21, 2026 (60 days from baseline) and December 21, 2026 (120 days), comparing against the August 21 baseline:

• Same 20 queries, 4 models (ChatGPT, Perplexity, Claude, Gemini)
• Score each response 0-3 based on whether YepIts.ai is mentioned and recommended
• Track: mention count, recommendation count, position in any lists

We'll publish results honestly on this page — including if YepIts.ai still doesn't appear in ChatGPT after 4 months. Honest reporting builds trust; hockey-stick charts built on 30 days of data don't.

What we'd do differently next time

• Audit the live build before adding more. The blog-not-bundled bug ate a day we could have spent on content. Always check that what you think is deployed is actually deployed.
• Don't link to profiles you don't own. The four broken sameAs URLs were actively harming YepIts.ai's entity signal. AI models downrank entities with contradictory or dead links.
• Pick a Twitter handle you control before generating entity content. We almost anchored YepIts.ai's entity to someone else's Twitter account.
• Wikidata is high-leverage for solo founders. One entry, ~20 minutes of work, ~3x citation boost per the Princeton research. Free.
• Comparison content targets ChatGPT's training gaps. The 3 new posts are all competitors ChatGPT named in the baseline. AI models extract from comparison tables aggressively — this is the highest-ROI content type for early GEO.

What this means for you

If you're considering AIOptim for your SaaS, here's what you should know:

• We'll do an audit first and tell you what's actually broken — not just pitch you on new content.
• We measure over months, not weeks. Citation growth is a slow game; we won't promise hockey-stick results in 30 days.
• We'll update this page publicly with real numbers as they come in. Good or bad.

Audit is free. 48-hour turnaround. No commitment.

What we're NOT claiming

To be transparent about the limits of this case study:

• We don't have data showing YepIts.ai is now appearing in AI recommendations — we just shipped the foundation, and citations take months to materialize
• We don't know if our methodology will replicate for your SaaS — GEO is highly category-dependent
• One client case study is anecdotal — we'll have more case studies as we accumulate work
• Our pricing reflects ongoing work, not one-time fixes

If you want a real test, ask us for a free audit. We'll tell you what's broken and what would actually move the needle for your category.