The playbooks, in full.

The problem. An AI language tutor competing against Duolingo, Babbel and Speak — brands with enormous domain authority and a decade of backlinks. Ranking for "language learning app" head terms was hopeless. But the buying question is never the head term. It is "is X better than Duolingo", "what should I use to learn German", "does ChatGPT work as a tutor".

What Klepha found. Gap analysis surfaced two clusters nobody had properly earned: head-to-head comparison queries against each incumbent, and beginner "where do I start with <language>" intent. High commercial intent, weak incumbent coverage — the incumbents write about their own product, not about the comparison.

What we built. A 14-page cluster: five head-to-head comparisons (vs Duolingo, Babbel, Speak, Langua, Praktika), two ranking pillars for "best" and "top 8" queries, two AI-versus-human angle pieces, and five learner-intent guides — all cross-linked, all shipped with Review, ItemList, FAQPage and BreadcrumbList schema so both Google and the answer engines can parse the verdict without guessing.

Why it works on AI engines too. Comparison pages with explicit, structured verdicts are exactly what an LLM quotes when someone asks it to compare tools. The same 14 pages that chase long-tail search also feed the answer engines the sentences they need.

The problem. A domain with no history, no backlinks and no citations, in a category where buyers increasingly ask an assistant instead of searching. Classic SEO advice — "build authority over 12 months" — is not a strategy when the answer engines are choosing vendors today.

What Klepha found. Running our own scans showed the uncomfortable truth: engines answered questions about SEO tooling by naming the same four incumbents, and none of the pages they cited were ours because we had none worth citing. The audit also flagged the mechanical blockers — pages the AI crawlers could not parse cleanly, missing structured data, no canonical answer to any question we wanted to own.

What we built. Ten deep technical guides — generative engine optimization, keyword gap analysis, schema markup, technical audits, on-page checklists and more — each written to be the quotable answer to one question, each with a visible FAQ block matched by FAQPage JSON-LD, breadcrumbs, and internal links into the product pages.

How we know it moves. Every prompt we care about is tracked weekly against ChatGPT, Gemini and Perplexity, with the answer text, the citations and the competitor mentions stored per scan — next to Search Console clicks and positions for the same topics.

The problem. Every day people describe your exact problem on Reddit and ask which tool solves it. Those threads rank in Google, get quoted by AI engines, and convert far better than an ad — but finding them manually means scrolling forever, and posting badly gets you banned.

Week 1 — listen properly. Set the keyword list that defines your niche, and let Klepha pull matching threads continuously. Every thread gets an AI relevance score, so a feed of hundreds collapses into the handful actually worth answering. Nothing is posted yet.

Week 2 — answer, do not advertise. Klepha drafts a reply for each high-scoring thread in the subreddit's register: answer the question first, disclose the affiliation, mention the product only where it genuinely fits. You review and edit every draft before it goes anywhere.

Week 3 — turn threads into pages. The recurring questions in those threads are keyword research that no tool can fake. Feed them into the content engine and publish the page that answers each one properly — the same question then earns you the thread reply and the search result.

Week 4 — measure both channels. Watch the tracked prompts and the Search Console queries for the topics you engaged. Community answers are also training data for the engines: threads where your brand is described accurately shift what assistants say about you.