The client runs an AI girlfriend fantasy platform — the most crowded new category in adult search, where a hundred near-identical products fight over the same keywords. Under NDA, as always. What we can show is a year of surgical repositioning: fewer impressions, more clicks, and an average position cut almost in half.
Read this one carefully — the impressions went down, and that was the plan.
A year ago this platform had what most AI-companion startups think they want: 1.17 million impressions a quarter. The problem was where those impressions lived — an average position of 28.4, deep on page three, being shown to users who never scroll. At a 5.6% click-through rate, the site was paying the full reputational cost of ranking (SafeSearch exposure, lookalike scraping, affiliate skimming) while collecting a fraction of the available clicks.
The AI girlfriend category compounds the problem. It's the most cloned vertical in adult search: dozens of products with near-identical landing pages, a query space where "ai girlfriend" head terms are fought over by venture-funded incumbents, and a long tail that shifts monthly as new fantasies and model names trend. Chasing impression volume here means ranking for everything shallowly. We chose the opposite.
The −28% impressions figure is the program's signature, not a loss. We de-indexed and consolidated the thin tag, parameter and near-duplicate pages that were generating page-three impressions by the hundreds of thousands, and collapsed the site's equity into the pages that could actually win. Fewer URLs, fewer query surfaces, more authority per surface. Google responded exactly as the model predicts: average position nearly halved, 28.4 → 14.5.
Head terms like "ai girlfriend" are a knife fight between funded incumbents — the export shows that battle still in progress. The wins came from the long tail: specific fantasy and niche-interest queries, format queries ("generator", "chat", "roleplay" variants), and trend-driven terms the big players are too slow to build for. The program shipped landing and editorial pages matched to those intents, each one engineered to answer the query better than a generic homepage ever could.
CTR climbed from 5.6% to 9.9% — at this scale that's tens of thousands of extra clicks from the same rankings. The levers: title and snippet systems rewritten per query intent, brand-SERP cleanup that pushed scrapers and affiliate lookalikes off page one for the brand's own name, and content structured so Google lifts the right answer into the snippet. In a category where users choose in seconds and mostly on mobile (78% of clicks), the snippet is the product demo.
The export shows clicks from 237 countries, with France, Spain and Latin America growing on their own locale pages — including top-position rankings on Spanish-language surfaces that didn't exist a year ago. The AI companion category is globally searched but English-only optimized; localized pages for localized fantasies is close to uncontested ground.
Note what's deliberately absent: an impressions brag. Most agencies would have hidden the −28% figure or spun it. We lead with it, because it's the difference between SEO as vanity and SEO as engineering — the site now ranks for a smaller, sharper universe of queries it can actually win, at positions users actually reach, with snippets they actually click. 83K clicks on 841K impressions beats 65.8K on 1.17M in every currency that matters: server costs, brand safety, and revenue per visit.
The next phase is already visible in the data: the position-4-to-15 band is crowded with pages one good quarter away from page-one dominance. That's where the compounding starts.
NDA — standard on every account, and especially strict in the AI companion category where clone operators monitor competitor marketing. What we can show is the Search Console comparison: 65.8K → 83K clicks, 1.17M → 841K impressions, CTR 5.6% → 9.9%, average position 28.4 → 14.5, across 237 countries.
Because position and CTR went up by far more. The old impressions were page-three noise: shown, never clicked, and costly in brand exposure. The new profile is 841K impressions the site actually competes for — which is why clicks rose 26% on 28% less visibility. We optimize for revenue per impression, not impression count.
It's the playbook we'd run from day one: resist the head-term fight, own the fantasy long tail, keep the index lean, treat snippets as the storefront. New products actually have an advantage — no legacy bloat to prune, so the "subtraction" phase is skipped.
Book a live audit and we open your Search Console comparison view together — which of your impressions earn clicks, which are page-three decoration, and what pruning would do to your average position. Run the free instant scan first if you want the snapshot ready.
GET A LIVE AUDIT WITH AN SEO EXPERTDISCREET · NDA ON REQUEST · SEE PRICING