# Losing half of Google traffic to AI Overviews > [!quote] > "We did not really recover our original programmatic SEO traffic from the original source. We created good practice, new sources in the new day and age that we are in." A podcast interview with SignLab's founder — "Duolingo, but for sign languages" — walking through losing >50% of Google clicks to AI Overviews and what actually brought traffic back. Named participants: the founder and their SEO consultant, David. ## Author's thesis The traffic that AI Overviews took was never recovered on its original terms. Recovery came from changing *what kind of content* was published and *which index it lived in* — not from optimizing the thing that got eaten. ## Key arguments 1. **What got extracted was a dictionary.** SignLab ran a programmatic SEO play: a free library of sign-language video clips, one page per sign, per language. A parent googling how to sign "stop" to a deaf toddler landed there and could download the app. Google began surfacing the video directly in the overview. Rankings and impressions held steady; **clicks went to near zero**. The founder's read: "Google kind of stole it." 2. **Funnel position predicted the damage.** Top-of-funnel content is the easiest for a model to regurgitate and gives the reader no reason to click. Expanding the programmatic play with *more* detail didn't help — "AI Overviews just takes your content and don't really give a lot back." What worked was deliberate middle- and bottom-of-funnel coverage, which the founder says still outperforms every other move they made. 3. **Republishing into a rival index recovered real traffic.** Roughly 4,000 clips were moved from their own storage onto YouTube — which the founder calls, after paid, their most important organic channel. This was operationally expensive: unverified-channel upload caps (~25–50/day, then ~250), a business-verification process brokered through their Google Ads rep to lift the cap to ~120/day, YouTube penalties on sub-3-second clips, and **12 separate channels** because sign languages are national (Libras Lab for Brazilian Sign Language, etc.) and the audience expects its own language, not English. The founder flags the same play would likely work on X (feeding Grok) and Instagram. 4. **AI was the thing that made 12 markets tractable.** The founder is explicit that agentic pipelines let them run per-country SEO across 12 domains in languages they don't read — "revolutionary in a positive way," set against the same technology eating their US and UK traffic. 5. **The 2025 detour is worth remembering.** Before settling on fundamentals they chased GEO tactics through a period the founder describes as "alchemy," including memory-injection tricks that both participants now name as LLM poisoning and pure black hat. It stopped working. ## Caveats Single company, single vertical, and an unusual one — short video assets and nationally partitioned audiences are exactly the shape that makes the YouTube move viable. Timeline details are hedged in the transcript ("don't quote me," "I might be off by a month") and at least one date is a transcription error. Their upload-cap relief came via an ads rep, which is not a path available to most sites. No numbers are given for the recovery, only direction and relative ranking of tactics. ## Links - [[3. Resources/Permanent Notes/Funnel Position Predicts Which Content AI Search Extracts]] — the diagnostic claim this note promotes - [[3. Resources/Permanent Notes/Republishing Into a Rival Index Recovers Traffic an Aggregator Intercepted]] — the remedy claim this note promotes - [[3. Resources/Permanent Notes/AI Search Inherits the Trust and Fragility of the Sites It Cites]] — the other side of the same transaction: the overview depends on the sites it is defunding - [[3. Resources/Permanent Notes/Search Visibility and Business Value Need Separate Metrics]] — evidence-for: the sharpest possible case, where rankings and impressions held perfectly steady while the business value went to zero - [[3. Resources/Permanent Notes/Perm · AI-generated content slop becomes SEO-weaponized when low-effort production is cheap]] — the LLM-poisoning detour is the same incentive one step further into black hat - [[3. Resources/Permanent Notes/A Metric Is Vanity Only Against Your Own Outcome Data]] — impressions were not vanity until clicks decoupled from them; the decoupling is the event - [[3. Resources/MOCs/DevRel MOC]] — applies-in: the MOC lists this note under Synthesis additions (2026-07-25). ## Source - [Reader: He Lost 50% of His Google Traffic to AI Overviews (Here's How He Recovered)](https://read.readwise.io/read/01kya7ehkbfad62npcq6s5hthq)