83 факта, 10 вопросов

Назад к отчёту

проверено 52частично 28не подтверждено 9

Что решает ранжирование и цитирование в 2026 году у Google, Яндекса и ИИ-ответов, и какие факты из вики неверны или устарели

Без цитаты: 0 · правила 12 · исследования 1 · опыт 4

провереноправилаdevelopers.google.com
There is no AI-text penalty as such; "scaled content abuse" applies "regardless of whether content is produced by automation, humans, or a combination".
провереноправилаdevelopers.google.com
E-E-A-T is a framework, not a scalar score; Google operationalises it through many signals and states there is no single helpful-content or E-E-A-T score.
провереноправилаgoogle.com
External links and brand mentions remain a core trust signal.
провереноправилаdevelopers.google.com
Originality is relative to the competition on that exact query, not absolute.
провереноправилаyandex.ru
Yandex's official quality frame is EPOS.
провереноправилаyandex.ru
Yandex states no fundamentally new requirements arise for Alisa answers because they are built on search results.
провереноправилаwebmaster.yandex.ru
Turbo pages are dead: announced 7 February 2025, off around April 2025, no replacement named, and Yandex says disabling them had no traffic effect.
провереноправилаyandex.ru
Five Yandex numbers are explicitly banned from plans as unverified: "commercial factors = 60–70% of weight", "95% of bought links ignored", "78% of legal queries geo-dependent", "Alisa cites exactly 5 sources", "Yandex has a YMYL mode like Google".
провереноправилаdevelopers.google.com
There is no separate AI index. Google AI Overviews and AI Mode draw from the normal Search index; eligibility is "indexed and snippet-eligible" with no additional technical requirements.
провереноправилаdevelopers.google.com
Google Search does not read or use llms.txt, and creating one has no positive or negative effect on AI Overviews or AI Mode visibility.
провереноправилаdevelopers.google.com
Schema.org markup is not required for Google's generative AI features, and there is no special schema needed.
провереноисследованияarxiv.org
Over 50% of desktop AI Overviews do not cite the #1 organic result.
провереноправилаdevelopers.google.com
Indexing is the gate, and Google promises it for no page.
частичноправилаdevelopers.google.com
There is no AI-text penalty as such; "scaled content abuse" applies "regardless of whether content is produced by automation, humans, or a combination".
частичноправилаdevelopers.google.com
External links and brand mentions remain a core trust signal.
частичноправилаyandex.ru
Originality is relative to the competition on that exact query, not absolute.
частичноправилаyandex.com
Turbo pages are dead: announced 7 February 2025, off around April 2025, no replacement named, and Yandex says disabling them had no traffic effect.
частичноправилаyandex.ru
Five Yandex numbers are explicitly banned from plans as unverified: "commercial factors = 60–70% of weight", "95% of bought links ignored", "78% of legal queries geo-dependent", "Alisa cites exactly 5 sources", "Yandex has a YMYL mode like Google".
частичноправилаdevelopers.google.com
There is no separate AI index. Google AI Overviews and AI Mode draw from the normal Search index; eligibility is "indexed and snippet-eligible" with no additional technical requirements.
частичноправилаdevelopers.google.com
Schema.org markup is not required for Google's generative AI features, and there is no special schema needed.
частичноисследованияarxiv.org
Over 50% of desktop AI Overviews do not cite the #1 organic result.
опытопытblog.bunnyhoneyclub.com
There is no AI-text penalty as such; "scaled content abuse" applies "regardless of whether content is produced by automation, humans, or a combination".
опытопытblog.bunnyhoneyclub.com
Originality is relative to the competition on that exact query, not absolute.
опытопытdev.to
External links and brand mentions remain a core trust signal.
опытопытseranking.com
There is no separate AI index. Google AI Overviews and AI Mode draw from the normal Search index; eligibility is "indexed and snippet-eligible" with no additional technical requirements.

Почему заброшенный futura.law обгоняет живой futura.inc в девять раз, и что из этого следует

Без цитаты: 0 · правила 2 · опыт 5

провереноправилаdevelopers.google.com
Google crawls a site as much as it wants to, and page quality sets that want.
провереноправилаdevelopers.google.com
Re-submitting the 13 overdue indexing tasks will not speed anything up.
опытопытhooshmand.net
A .net to .com move on 1 July 2022 with server-level 301s, the Change of Address tool and a verified new property still cut Google traffic from 1,000+ sessions a day to 50 and took 1.5 years to recover, so redirects did not carry the link value over on the expected timescale.
опытопытrapidfireweb.com
A root-domain change in February 2024 that kept the URL structure and used platform redirects still cost 72% of clicks and 85% of impressions comparing January 2024 with mid-February to mid-March 2024, and the owner names asking the largest linking sites to repoint their links as a step she would not skip.
опытопытmedium.com
A staged move from sunnyvalley.io to zenarmor.com after a lost core update, run over three months with cross-domain canonicals, matched brand mentions on third-party sites such as G2, Capterra and Wikipedia, and only then a full redirect, produced a 110% traffic increase, and the new domain outranked the old one despite having no links of its own.
опытопытgsqi.com
On one client domain change the Crawl Stats report for the old domain showed 4% of Googlebot requests ending in 404 instead of 301, which is the measurable gap where inbound links stop passing, against a correct migration where 100% of requests redirect.
опытопытjoost.blog
The 2013 guardian.co.uk to theguardian.com move, run as engineering work with SEO embedded from day one and page-by-page redirects, still showed record traffic six months later, against Topshop's wildcard redirect into ASOS in 2021 which lost the visibility for good.

Почему машинные статьи не попадают в индекс Google и что чинит индексацию

Без цитаты: 0 · правила 4 · опыт 7

провереноправилаdevelopers.google.com
A page enters AI Overviews and AI Mode only if it is indexed and snippet-eligible, and the site must also be included in one Search Console setting. The wiki belief "no additional technical requirements" is now out of date.
провереноправилаyandex.ru
Yandex names a catalogue without its own texts as a low-quality marker, which is the shape of the 419 format pages.
провереноправилаdevelopers.google.com
Zero indexed articles means zero AI Overview and AI Mode eligibility, because there is no separate AI index.
частичноправилаdevelopers.google.com
Templated pages differing only by a variable match Google's doorway definition; every short page needs data that exists nowhere else.
не подтвержденоправилаsupport.google.com
Templated pages differing only by a variable match Google's doorway definition; every short page needs data that exists nowhere else.
опытопытdev.to
Templated pages differing only by a variable match Google's doorway definition; every short page needs data that exists nowhere else.
опытопытdev.to
On a five-month-old 5,800-page site Google indexed 428 of 1,432 generated timezone pages (29.9%) and rejected 647, while all 2,290 handwritten recipes sat in "discovered, currently not indexed" and were never crawled, so the block was crawl allowance on a domain with 0 external links, not text quality.
опытопытdev.to
On a dropped domain caught in March 2025, 1 of 57 pages was indexed 13 months later (38 discovered, 17 crawled and not indexed) after clean JSON-LD, GSC URL removals of the prior owner's paths, PSI 78 to 98 and TTFB 1.6s to 0.64s, and the indexed count stayed at 1.
опытопытblog.bunnyhoneyclub.com
Two client sites publishing several hundred machine-written posts lost 81% and 74% of organic traffic in a week with 600+ posts deindexed; stopping publication, deleting the worst pages with 410 Gone and rewriting the middle tier by hand brought both back to within 20% of pre-hit traffic two core updates later.
опытопытseranking.com
2,000 one-click machine-written articles published on 20 brand-new domains on 5-6 November 2024 with only a sitemap submitted reached 70.95% indexing (1,419 pages) in 36 days, so indexing was not the failure; the share of pages in the top 100 then fell from 28% to 3% around month three and had not recovered after 16 months.
опытопытblog.naibabiji.com
An article drafted by a model and then reviewed and edited by hand, published 26 July 2026, was indexed by Google within 9 hours with screenshot proof, and the author reports all published content on the site indexed.
опытопытmedium.com
An article that stayed unindexed for nearly two weeks was indexed within 48 hours after the author rewrote the title around a searched keyword, put the keyword in the first 100 words, replaced headings with search questions, edited the page 48 hours after publishing and requested indexing in Search Console.

Ссылки и упоминания: что работает для маленького сайта без покупки размещений

Без цитаты: 0 · правила 3 · исследования 1 · опыт 4

провереноисследованияarxiv.org
The Princeton GEO paper (KDD 2024) reports that adding statistics, quotations and source citations raised generative visibility by up to ~40%.
провереноправилаdevelopers.google.com
The only sanction risk is undisclosed paid links; paid placements marked `rel="sponsored"` are fine.
провереноправилаgoogle.com
Links from prominent sites stay a documented quality signal; unlinked brand mentions are not, so belief b06 is partial.
частичноисследованияarxiv.org
The Princeton GEO paper (KDD 2024) reports that adding statistics, quotations and source citations raised generative visibility by up to ~40%.
частичноправилаdevelopers.google.com
You cannot confirm a backlink by fetching the page: Cloudflare-fronted sites serve a 5KB stub to a script while DataForSEO crawled the real 500KB page, and JS-rendered profiles hide the link from raw HTML. Both directions give false answers; only a human opening the page settles it.
не подтвержденоправилаdevelopers.google.com
The only sanction risk is undisclosed paid links; paid placements marked `rel="sponsored"` are fine.
опытопытgsqi.com
The only sanction risk is undisclosed paid links; paid placements marked `rel="sponsored"` are fine.
опытопытeditorialge.com
94% of posts have zero external links, 1.3% of posts earn 75% of shares, and data studies and free tools earn about 2× the referring domains of guides.
опытопытeditorialge.com
A first batch of 50 link requests to editors got 23 replies and no links, while one how-to page published in April 2023 earned 79 referring domains over 40 months with zero outreach emails, nothing for the first 18 weeks and then 3 to 4 new domains a month.
опытопытx.com
Instead of paid placements, a solo founder built a free startup leaderboard and a bot that posts other founders' Stripe payments; the bot reached 700 followers in a few days, added 56% to IndiePage revenue and pushes thousands of impressions a day to the listed startups.

Какие инструменты нужны одному человеку для ключевых слов, аудита, позиций, ссылок и ИИ-цитирований, и от каких отказаться

Без цитаты: 0 · правила 8

провереноправилаwebmaster.yandex.ru
Yandex Webmaster has a free official tool "Видимость сайта в Алисе AI" under Эффективность, showing share of voice.
провереноправилаwebmaster.yandex.com
Semrush cannot see Yandex; Serpstat, Topvisor and Rush Analytics can; your own site is free through Yandex Webmaster and Wordstat.
провереноправилаyandex.com
IndexNow is supported by Bing, Yandex, Naver and Seznam but not Google; ChatGPT Search draws on Bing's index, so IndexNow is the cheapest lever for both Yandex indexing and ChatGPT freshness.
провереноправилаhelp.openai.com
ChatGPT citation depends on one crawler, not on any file or markup.
провереноправилаdevelopers.google.com
The `llms.txt` route shipped on 2026-08-27 does nothing for Google, so the wiki's AEO-lever belief is refuted.
провереноправилаyandex.ru
Yandex Webmaster reports the site's share of Alisa AI answers, the only first-party AI-citation metric this site has.
провереноправилаdevelopers.google.com
Google names the exact failure of the ~500 machine-written pages judged "technically correct but generic".
провереноправилаyandex.com
IndexNow reaches Yandex but no Google page in the corpus mentions it, so it is a Yandex lever only.
частичноправилаwebmaster.yandex.ru
Semrush cannot see Yandex; Serpstat, Topvisor and Rush Analytics can; your own site is free through Yandex Webmaster and Wordstat.
частичноправилаblogs.bing.com
IndexNow is supported by Bing, Yandex, Naver and Seznam but not Google; ChatGPT Search draws on Bing's index, so IndexNow is the cheapest lever for both Yandex indexing and ChatGPT freshness.

Факты из вики, проверены

Без цитаты: 0 · правила 17 · исследования 6 · опыт 8

провереноправилаdevelopers.google.com
Quality is site-wide, not per page; the Helpful Content System merged into core ranking in March 2024, so weak pages drag the whole domain.
провереноправилаblog.google
Google projected the March 2024 update would cut low-quality unoriginal content by 40%, revised to 45% after rollout.
провереноправилаsupport.google.com
Rankings take 3–6 months; report position and click deltas, not vibes.
провереноправилаdevelopers.google.com
Schema does not cause citations: the Ahrefs causal arm matched 1,885 newly-schema'd pages against ~4,000 controls and measured +2.4% AI Mode, +2.2% ChatGPT, −4.6% AI Overviews.
провереноправилаhelp.openai.com
ChatGPT Search uses two independent crawlers.
провереноправилаdocs.perplexity.ai
Perplexity runs PerplexityBot (index) and Perplexity-User (live user fetches); PerplexityBot's documented sole purpose is to surface and link sites, not to train models.
провереноправилаsupport.claude.com
Anthropic runs ClaudeBot (training), Claude-User (user-initiated fetch, blocking it "may reduce your site's visibility for user-directed web search") and Claude-SearchBot (search index).
провереноправилаdevelopers.google.com
llms.txt is ignored in practice: Ahrefs found 97% of llms.txt files got zero requests in May 2026 across 137K domains, and four independent server-log studies covering 68,759 AI-bot requests found zero or single-digit fetches by GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot; SE Ranking found no relationship between llms.txt and citation frequency across ~300,000 domains.
провереноправилаdevelopers.google.com
Google restricted FAQPage rich results for commercial sites in August 2023, so FAQ schema is not an AI-visibility lever, though a visible FAQ section still helps chunk extraction.
частичноправилаdevelopers.google.com
Quality is site-wide, not per page; the Helpful Content System merged into core ranking in March 2024, so weak pages drag the whole domain.
частичноправилаservices.google.com
Legal content is YMYL and trust signals weigh especially heavily there.
частичноправилаdevelopers.google.com
Rankings take 3–6 months; report position and click deltas, not vibes.
частичноправилаwebmaster.yandex.ru
Alisa issues several internal search queries and then picks the highest-quality pages.
частичноправилаyandex.ru
Behaviour outweighs links in Yandex: dwell time, direct-visit share, bookmarks, repeat visits; the ПФ filter punishes click manipulation; Minusinsk (2015) devalued bought links.
частичноправилаyandex.ru
Alisa citations from outside the Yandex top 10 fell from 40% to 10% in six months, so you generally need organic top 5.
частичноправилаyandex.com
Regional binding is a ranking factor, and choosing "Moscow and oblast" instead of "Moscow" can cost 20–40 positions.
частичноисследованияarxiv.org
Schema does not cause citations: the Ahrefs causal arm matched 1,885 newly-schema'd pages against ~4,000 controls and measured +2.4% AI Mode, +2.2% ChatGPT, −4.6% AI Overviews.
частичноправилаdevelopers.openai.com
ChatGPT Search uses two independent crawlers.
частичноправилаdocs.anthropic.com
Claude fetches live at question time and has no index, so JS-only SPAs are invisible to it.
частичноисследованияarxiv.org
Being indexed is mandatory, being #1 is not: only 16.7% of AI Overview citations come from the organic top 10 and 54.5% rank organically at all; AI Overviews cite 2–7 sources.
частичноправилаdevelopers.google.com
Branded web mentions correlate 0.664 with AI Overview visibility, far stronger than backlinks or Domain Rating.
частичноправилаdevelopers.google.com
llms.txt is ignored in practice: Ahrefs found 97% of llms.txt files got zero requests in May 2026 across 137K domains, and four independent server-log studies covering 68,759 AI-bot requests found zero or single-digit fetches by GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot; SE Ranking found no relationship between llms.txt and citation frequency across ~300,000 domains.
частичноисследованияarxiv.org
Engine mixes differ: Gemini favours first-party sites, Perplexity favours Reddit/YouTube/fresh, Claude is the most conservative citer.
частичноправилаdevelopers.google.com
Google restricted FAQPage rich results for commercial sites in August 2023, so FAQ schema is not an AI-visibility lever, though a visible FAQ section still helps chunk extraction.
не подтвержденоправилаdevelopers.google.com
Quality is site-wide, not per page; the Helpful Content System merged into core ranking in March 2024, so weak pages drag the whole domain.
не подтвержденоправилаsupport.claude.com
Claude fetches live at question time and has no index, so JS-only SPAs are invisible to it.
не подтвержденоисследованияarxiv.org
Being indexed is mandatory, being #1 is not: only 16.7% of AI Overview citations come from the organic top 10 and 54.5% rank organically at all; AI Overviews cite 2–7 sources.
не подтвержденоисследованияarxiv.org
Heavy Reddit brand mentions correlate with 3.9× more ChatGPT citations, visible in 3–6 weeks; Perplexity's top citation source is Reddit at 46.7%, then YouTube at 13.9%.
не подтвержденоправилаdocs.perplexity.ai
llms.txt is ignored in practice: Ahrefs found 97% of llms.txt files got zero requests in May 2026 across 137K domains, and four independent server-log studies covering 68,759 AI-bot requests found zero or single-digit fetches by GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot; SE Ranking found no relationship between llms.txt and citation frequency across ~300,000 domains.
не подтвержденоисследованияarxiv.org
AI-cited content is roughly 25% fresher than SERP winners, and Perplexity cites within hours (82% of citations under 30 days old).
не подтвержденоисследованияarxiv.org
Listicles as an AI-visibility play decay about 30% month over month in ChatGPT.
опытопытblog.bunnyhoneyclub.com
Quality is site-wide, not per page; the Helpful Content System merged into core ranking in March 2024, so weak pages drag the whole domain.
опытопытblog.bunnyhoneyclub.com
Rankings take 3–6 months; report position and click deltas, not vibes.
опытопытcentori.io
Behaviour outweighs links in Yandex: dwell time, direct-visit share, bookmarks, repeat visits; the ПФ filter punishes click manipulation; Minusinsk (2015) devalued bought links.
опытопытcentori.io
Regional binding is a ranking factor, and choosing "Moscow and oblast" instead of "Moscow" can cost 20–40 positions.
опытопытcjr.org
ChatGPT Search uses two independent crawlers.
опытопытcjr.org
Perplexity runs PerplexityBot (index) and Perplexity-User (live user fetches); PerplexityBot's documented sole purpose is to surface and link sites, not to train models.
опытопытseranking.com
Being indexed is mandatory, being #1 is not: only 16.7% of AI Overview citations come from the organic top 10 and 54.5% rank organically at all; AI Overviews cite 2–7 sources.
опытопытranki.io
Google restricted FAQPage rich results for commercial sites in August 2023, so FAQ schema is not an AI-visibility lever, though a visible FAQ section still helps chunk extraction.

Вопрос Q3

Без цитаты: 0 · опыт 3

опытопытgsqi.com
A 20K-URL YMYL site carrying sections of machine-written text was fully deindexed in April 2026, but the cause was a hijacked www subdomain spiking to nearly 12K clicks a day on gambling queries; closing the DNS hole and filing a reconsideration request restored the index within about two days.
опытопытsearchpilot.com
A controlled split test on a travel site's listing pages added machine-written destination copy rendered server-side as well as client-side, so it sat in the HTML, and organic sessions rose an estimated 18%.
опытопытbree-sharp.com
A March 2026 Search Console audit found three 404 URLs still ranking, the best at position 6.32 with 313 impressions a month and 472 combined; rebuilding the articles at the exact same paths rather than redirecting got them reindexed the same session.

Сайт: futura

Без цитаты: 0 · правила 5

провереноправилаdevelopers.google.com
The 656-URL schema audit fixes rich results, not AI citations; the wiki belief that schema does not cause citations is confirmed in direction, but its Ahrefs numbers have no trusted source.
провереноправилаservices.google.com
The wiki belief "legal content is YMYL" is partial: Google defines YMYL by health, financial stability, safety and welfare, and never names legal content.
провереноправилаyandex.ru
Yandex, unlike Google, names law directly as a field where author expertise decides ranking.
провереноправилаyandex.ru
Yandex states that AI text without an author's added meaning fails its originality aspect.
провереноправилаyandex.ru
Alisa AI answers draw on pages that already rank high, so the Alisa visibility tool reports nothing until positions exist.

Сайт: improv-tovarisch

Без цитаты: 0 · правила 3

провереноправилаdevelopers.google.com
Google's bar is value relative to the competing results, not data that exists nowhere else, so the wiki's stricter rule (belief b08) is partial.
провереноправилаdevelopers.google.com
AI Overviews and AI Mode need no work beyond ordinary indexing and snippets.
провереноправилаyandex.ru
Yandex applies one frame, ЭПОС, to ranking and to Alisa AI answers, so the RU plan needs no separate AEO track.

Сайт: starlie

Без цитаты: 0 · правила 5 · исследования 1

провереноправилаdevelopers.google.com
Google Search ignores `llms.txt`, so the deployed file changes nothing for Google. The wiki belief is confirmed.
провереноправилаdevelopers.google.com
A new site usually stays unindexed because it is new, and a sitemap does not guarantee indexing.
провереноправилаdevelopers.google.com
Crawling of the four new URLs takes days to weeks, and repeated indexing requests do not speed it up.
провереноправилаdevelopers.google.com
Google tells site owners not to publish content that a generative model could produce, which is exactly how the four pages were built.
провереноправилаgoogle.com
Links from prominent sites remain a quality and trust factor. The wiki's added claim about brand mentions has no trusted source.
провереноисследованияarxiv.org
AI citation does not follow the organic top 10, so Starlie can be cited before it ranks. The wiki numbers 16.7%, 54.5% and "2-7 sources" are refuted; measured AI Overviews cite about 12 sources.

Вопросы: 5 · Источники: 90 · Цитаты: 194 · Факты: 83 · Ответы: 5 · $0.30