Video summary

Попади в ТОП-1 Google и Яндекс. SEO тексты Claude работают? Инструкция и промты

Main summary

Key takeaways

Educational

Main ideas / lessons

  • SEO in 2026 is less forgiving of “empty” autogenerated content.

    • Search engines (especially via behavioral signals and intent satisfaction) are better at detecting low-value texts.
    • Simply generating and publishing text is unlikely to move pages to the top positions anymore.

  • User behavior signals are a core ranking factor (organic traffic, not “cheating”).

    • Promotion depends heavily on satisfying intent and improving behavioral metrics.
    • Focus on site usability and engagement, aiming for:
      • higher average viewing time
      • lower bounce rate
    • Usability matters across desktop, mobile, and tablet layouts.

  • Content must add real value; “similar/duplicate” autogenerated content degrades performance.

    • Content without unique insight, expert opinion, or a distinct professional message tends to underperform even if keyword-optimized.
    • The “value of content is growing every day.”

  • Generated texts can work, but typically require a human layer and better prompting.

    • Position: use autogenerated text as raw material, but copywriter post-processing is a must.
    • Copywriter tasks include:
      • editing paragraphs and phrases
      • adding genuinely unique information
      • removing or correcting issues like unnatural keyword repetition

  • For generative/AI-driven search (“GEO/GeE”), credibility and citations matter.

    • The approach is expected to improve visibility not only in classical results, but also in:
      • Google AI Overviews
      • Yandex results (including Alice / Yandex Neural)
    • For highly technical content, it’s recommended to:
      • cite sources
      • mention where information was obtained (manuals, patents, research works)
      • optionally list resources without relying on external link spam

  • Mass, large-scale auto-generation has risks (especially in Google).

    • Described experiment:
      • auto-generated ~1,000 articles via a WordPress plugin
      • Yandex indexed them first, then gradually removed pages
      • Google initially indexed them, then removed them all later, causing traffic loss
    • Conclusion: mass autogenerated/templated sites may be deindexed or fail to index long-term—prefer a text-by-text approach.

Methodology / instructions (detailed)

A) What to do to improve SEO outcomes with AI text generation

  • Don’t rely on primitive mass generation tools without post-processing.
  • Generate content in a structured way, not just “write an SEO text about X.”
  • Use complex prompts with context and constraints, such as:
    • role/persona of the model (e.g., SEO copywriter/content editor)
    • strict instruction: create one unique article, not 10–40 at once
    • technical spec requirements:
      • generate title/description manually first, then feed them into the process
      • heading structure: H1–H3 only (their practice avoids going beyond H3)
      • number of headings/sections
      • character count/length
      • paragraph structure and sentence length (avoid both very long and very short)
    • keyword handling rules:
      • provide a list of key phrases
      • specify where/how they should appear (based on semantic core + clustering)
    • writing style constraints:
      • define a project Tone of Voice with the client team
      • optionally control stylistic concentration (example: “style at 10–15%”)

B) Required human post-processing (copywriter role)

After neural-network generation, the copywriter should:

  • edit and rewrite parts that sound generic or repetitive
  • add unique “insider” information from the company
  • correct keyword overspamming
  • improve readability so the article “reads like an article,” not a template

C) Strengthen perceived expertise and uniqueness

  • Provide the model insider/company-specific statistics and analytics, e.g.:
    • attendance numbers
    • participants
    • number of events
    • year-over-year metrics
  • Include the author’s attitude/opinion:
    • ideally from an expert in the company
    • if the model can’t express attitude well, do a short interview and insert a quote

D) Add engagement + interaction features on the page

To increase engagement, enrich pages with interactive blocks such as:

  • surveys
  • quizzes
  • complex forms
  • other functional blocks that encourage clicking/interacting

For longreads:

  • include a table of contents at the beginning
  • use anchor links to sections to increase scroll depth and reading completion

E) Site technical/usability factors to prioritize (on all devices)

Evaluate and optimize:

  • mobile UX (emphasized as top priority for B2C; also important for B2B)
  • tablet versions
  • avoid usability problems across platforms

They also suggest ensuring desktop + mobile + tablet versions perform well for engagement metrics.

F) Indexing/troubleshooting procedure when content doesn’t index

  • If Google sees the page but delays indexing:

    • wait about 2–3 weeks
    • use forced indexing
    • if Google still refuses:
      • re-optimize / rewrite and recompile the text
      • resubmit for re-indexing

  • If the page gets indexed but drops out later:

    • continue monitoring
    • strengthen external credibility signals:
      • add/strengthen external links (the speaker states “buy links/link building” as a working strategy)

  • Special indicator mentioned:

    • Yandex indexes but Google doesn’t → likely content quality issues (in their view)

Tools / system described (workflow chain)

The speaker describes an internal multi-model pipeline for creating long-form SEO articles:

  1. Step 1: GPT Chat

    • creates the article structure
    • defines paragraph lengths and number/coverage of headings for a longread

  2. Step 2: “GM”

    • converts the structure into bullet points / main-message outlines

  3. Step 3: “GMI / Gini”

    • expands bullet points into paragraphs
    • aims to keep reading engaged through the full text

  4. Step 4: “KLOD” / Claude

    • converts into the final coherent text

  5. Step 5: copywriter post-processing

    • final edits + unique insights + author attitude + keyword cleanup

They emphasize that skipping this “complex linking” makes outputs more “typical,” increasing the risk that Google crawls but does not index due to perceived low quality.

Speaker’s claims about “is it allowed / will you be banned”

  • The speaker frames the main risk as search ranking/indexing behavior, not necessarily an explicit ban.
  • They argue:
    • mass autogenerated content is generally not indexed correctly
    • they reference Google documentation as support, while noting they can’t guarantee 100% without tests
  • Recommended stance:
    • use AI generation, but with constraints + post-processing

Calls to action (business offer)

  • The speaker offers a free SEO and “Geo” audit.
    • Contact via website form / Telegram / links in description
    • Audit includes deep site analysis and identification of growth points/problem areas

Speakers / sources featured

  • Ilya Kuzchenkov — host/speaker
  • Webzusus / WebGUS — the speaker’s team/brand
  • Google — search engine (documentation referenced; indexing behavior discussed)
  • Yandex — search engine (indexing behavior and Alice/Yandex Neural mentioned)
  • GPT / GPT Chat — AI model/tool mentioned
  • Neural networks / LLMs — general AI systems mentioned
  • Claude — referenced in the video title and tools context
  • Telegram — platform where the full prompt file is said to be shared
  • WordPress — mentioned as part of the mass-generation experiment workflow
  • Conference/owner of a similar service — unnamed third party mentioned indirectly

Original video