Anthropic обвинила DeepSeek, Moonlight и Alibaba в дистиляции

Статья о том, как китайские компании массово выкачивают ответы западных нейросеток, чтобы дообучать свои модели. Anthropic говорит о 24 000 фейковых аккаунтов и 16 млн запросов к Claude. Вывод простой: рынок ИИ входит в фазу жёсткого копирования, а китайские модели Qwen, Kimi и DeepSeek могут быстро догонять лидеров дешевле.

➡️ Читайте на сайте: https://aff.top/blog/anthropic-obvinila-deepseek-moonlight-i-alibaba-v-distiliacii

🧠 Ещё больше инсайтов → в канале AFF.top

Auditing one page for expertise signals a rater can verify

The question: strip away the byline and the bio box — does the body of this page itself demonstrate expertise, or does it merely assert it elsewhere on the page?

The QRG distinguishes the expertise of the creator from the quality of the content, but in practice raters read the content for evidence the creator knows the subject. An in-body audit:

— Terminology precision: does the text use the field's actual vocabulary correctly, or generic synonyms a non-expert would reach for?
— Appropriate caveats: experts qualify claims where the evidence is genuinely uncertain. Uniform confidence across a complex topic is a non-expert tell.
— Non-obvious detail: the page surfaces things only practice teaches, not what the top three competing pages already say.
— Correct hierarchy: an expert foregrounds what matters and de-emphasizes the trivial; a generalist gives everything equal weight.
— Accurate framing of disputes: experts represent the state of debate fairly rather than flattening it to one answer.

Score the page on how much of the expertise evidence lives in the body versus the bio. Bio-only expertise is the weaker configuration.

Caveat: in-body expertise signals are read most reliably by human raters; the degree to which algorithms infer expertise from linguistic features is unestablished, and surface fluency can mimic several of these markers. Treat this as a quality discipline, not a ranking lever.

What we still don't know: whether any production ranking system distinguishes genuine domain expertise from competent generalist writing on linguistic grounds alone — a question even the QRG resolves by asking humans, not machines.

A contact-information completeness pass for a transactional site

The question: for a site that takes money or gives consequential advice, is your contact and accountability information at the level the QRG treats as a trust requirement?

The guidelines are specific that the amount of contact information appropriate depends on the site's type and that YMYL and transactional sites need more. A completeness pass:

— A genuine, monitored contact pathway: an email or form that reaches a human, and where the business model warrants it, a phone number and address.
— Clear identification of the responsible legal entity, matching what business registries and payment processors show.
— Policies a consequential transaction implies: returns, refunds, privacy, terms — present, specific, and not boilerplate that contradicts actual practice.
— Customer-service accessibility proportionate to risk: a site selling supplements needs more reachability than a hobby blog.
— Consistency between the contact details on-site and those an independent search surfaces; a mismatch is itself a trust flag.

Caveat: contact completeness is a trust hygiene factor, not a ranking amplifier. The QRG frames missing contact information as a reason to lower trust for sites that should have it — its presence prevents a deduction more than it earns a boost. A fully detailed contact page on a fraudulent site does not redeem the fraud.

What we still don't know: how, or whether, algorithmic systems detect contact-information adequacy at scale, versus this being primarily a human-rater trust input that reaches ranking only indirectly.

Тем временем подстилка коричневых Кардиналов и лично Кустова главред Максим Огненный завёл собственный канал, где, наверное, опять будет писать стихи и анонсировать вьюхи с ме**дроновыми наркоманами.🤡🤡

Чем вообще известен этот персонаж? Собсна, только порцией отборного кринжа, например, не так давно он выебывался на Иванова в "Письмах кардинала", но недожал и тема осталась нераскрытой. ЕЮ заслужил даже высокоинтеллектуальные выпады, которые тот не понял в силу врождённого аутизма:

Высокохудожественная лексика героя, демонстрируемая им не только на своем канале, но и на любой публичной площадке, настолько пленит любого слушателя, что мало кто может стать собеседником жертвы во второй раз.

Нихуя не разбираясь в аффилке, он на серьёзных щах брал интервью у Мефмедова и пытался построить серьёзный диалог с объёбанной ракетой. Из остальных достижений только нытье в чатиках: персонаж настолько невзрачный, что даже хуесосить его не так интересно.¯\_(ツ)_/¯

А теперь Огненный завёл собственный канал, где собрался выебать всех, но, походу, ебать будет лишь подписоту своей графоманией. Похожей хуйнёй занимался и оунер ФБ Киллы, который так любит рассуждать про экономику и политику. Оно и понятно: что у коричневых, что у киллы, что у партнёркина — один инвестор и одна крыша, поэтому подопечные синхронно занимаются бестолковой хуйнёй. Кому не похуй, подписывайтесь, будет интересно (нет): https://t.me/+dr240TeLtPc2Nzdk

В арбитраже деньги есть, но только у тех, кто с LuckyCards 💵

A refresh protocol that updates substance, not just dates

The question: a YMYL page is aging — what refresh sequence restores its trustworthiness rather than performing freshness by editing the timestamp?

The QRG treats outdated information on YMYL topics as a real trust problem, and Google has separately discouraged fake freshness signals. A substance-first protocol:

— Re-verify every factual claim against current primary sources. Guidelines change; a recommendation that was correct three years ago may now be wrong, which is worse than merely old.
— Identify claims that have been superseded and replace them, noting the change where transparency serves the reader.
— Re-confirm that cited sources still exist, still say what you claimed, and haven't themselves been retracted or updated.
— Have a credentialed reviewer re-review YMYL substance and re-attribute the review with a current date and their name.
— Only after substantive revision, update the visible "last reviewed" date — and only if the date reflects real review.

Caveat: changing a date without changing content is the failure mode this protocol exists to prevent. There is no evidence that timestamp manipulation produces durable ranking benefit, and on YMYL topics a fluent-but-stale page is a trust liability whatever its displayed date.

What we still don't know: how Google's systems weight content freshness against established page authority for queries where currency genuinely matters — the trade-off between a recently updated newcomer and a long-established but slightly dated authority is not publicly specified.

Reputation lives off your domain, and raters are told to go find it

The question: when a quality rater assesses your trustworthiness, where do they look?

Not primarily at your own pages. The QRG instructs raters to perform reputation research — to leave the website and search for independent, third-party information about the creator and the site. The guidelines name specific source types: independent reviews, references, news articles, expert recommendations, and for some entities, Wikipedia or industry-specific reputation sources.

The instruction is asymmetric. A site saying 'we are the trusted leader' counts for little. The guidelines explicitly tell raters that the website's own claims about itself are not reputation; reputation is what independent sources say. A small number of credible, specific third-party references outweighs volumes of self-description.

The guidelines also instruct raters to weigh the credibility of the reviewers themselves, and to be skeptical of suspiciously uniform praise — a signal that maps onto the reality of bought and seeded reviews.

For practitioners, the operational consequence is that off-page reputation work — earned mentions, being cited by credible third parties, accurate listings — is doing the heavy lifting that on-page tweaks cannot.

Caveat: rater reputation research informs the evaluation of the systems, not a direct per-URL input. The systems approximate it with their own signals; the two are correlated, not identical.

What we still don't know: the threshold at which a thin or absent external reputation actively suppresses a site versus merely failing to help it.

Antropic раскрыла ферму с нейронными дейтинг-моделями

Anthropic раскрыла кейс китайской студии, которая запустила 20 дейтинг-приложений с нейропрофилями для удержания пользователей. Вместо стандартного слива дейтинг-трафика на чужие смартлинки, вебмастерам стоит присмотреться к разработке собственных LLM-сервисов. Затраты на токены и подключение платежных решений окупаются за счет прямого контроля над монетизацией и забора всей маржи рекламодателя.

➡️ Читайте на сайте: https://aff.top/blog/antropic-raskryla-fermu-s-neironnymi-deiting-modeliami

🧠 Ещё больше инсайтов → в канале AFF.top

Разработчик игры слил $220 в Google ads на установки ботами

Статья описывает кейс разработчика, потерявшего бюджет в Google Ads из-за фрода и специфики атрибуции. Платформа засчитала конверсии, которые не подтвердились в Google Play Console, так как алгоритм учитывает установки без прямой связи с кликом. Главный вывод: Google Ads может наливать ботов не меньше пуш-сетей. При работе за инсталы важно жестко контролировать качество трафика и сверять аналитику с бэкендом, в то время как Facebook на данный мо…

➡️ Читайте на сайте: https://aff.top/blog/razrabotchik-igry-slil-220-v-google-ads-na-ustanovki-botami

🧠 Ещё больше инсайтов → в канале AFF.top

Google расширил Data Manager

Google интегрировал инструмент Data Manager в GA и DV360 для удобной передачи first-party данных, что дает рост ROAS до 26%. Для арбитража трафика прямого применения у офлайн-данных нет, однако инструмент можно использовать для манипуляции алгоритмами: отправка синтетических конверсий через API поможет скорректировать оптимизацию и направить автостратегии на поиск нужной аудитории.

➡️ Читайте на сайте: https://aff.top/blog/google-rasshiril-data-manager

🧠 Ещё больше инсайтов → в канале AFF.top

About, Contact, and policy pages are scored as trust infrastructure

The question: do 'boring' pages — About, Contact, editorial policy, terms — affect quality assessment, or are they decoration?

The guidelines treat them as load-bearing trust infrastructure, not decoration. For YMYL and transactional sites the QRG instructs raters to look for satisfying information about who is responsible for the website and its content, and adequate customer service or contact information. Their absence on a site that handles money or gives consequential advice is a documented driver toward the lowest ratings.

The mechanism is reputation and accountability. A findable, specific owner — a real organization, a real address, real contact paths — makes the site accountable for its claims. Anonymity on a YMYL site removes that accountability and, per the guidelines, the rater treats inability to determine who is responsible as a serious trust deficit.

The specificity matters. A generic contact form on a store that takes credit cards is weaker than a named company, physical address, and responsive support channel. An editorial or correction policy on a publication signals a process for accountability that a bare byline does not.

For affiliate operators, this is among the cheapest high-confidence work available: honest ownership disclosure, real contact, clear policies. It does little for 'Highest' but a great deal to clear the 'Lowest' triggers.

Caveat: thresholds differ by site type. A personal hobby blog is not held to the customer-service standard of an e-commerce store.

What we still don't know: how directly the live systems read these pages versus inferring accountability from other signals.

Vetting a guest contributor before their byline carries your trust

The question: a guest author wants to publish on your site — what verification protects the reputation a rater attributes to your whole domain?

The QRG's spam discussion flags third-party and guest content produced primarily to exploit a host's reputation. A guest byline also imports the guest's own reputation onto your pages, in both directions. A vetting checklist:

— Verify the person is real and the claimed credentials check out against issuing bodies — the same chain you'd apply to staff authors.
— Run the reputation search a rater would run on the individual; a guest with a negative independent reputation transfers risk to your domain.
— Confirm genuine subject expertise through prior verifiable work, not a polished pitch.
— Establish that the content serves your readers, not the guest's link-building agenda; the QRG and link-spam policies both target content placed mainly to pass links.
— Apply your standard editorial review and fact-check; a guest byline does not exempt the page from your accuracy bar.
— Keep accountability clear on the page: who wrote it and that your site stands behind it.

Caveat: the line between a legitimate expert contribution and reputation-exploiting guest content is drawn by intent and quality, which the QRG describes qualitatively. A single excellent guest piece is not abuse; a stream of thin third-party posts placed for links is.

What we still don't know: the volume and pattern thresholds at which legitimate guest contribution reads to Google's systems as a third-party-content abuse signal — Google has described the abuse but not quantified the boundary.

Xchat удалён из App Store и Google play

Google Play и App Store удалили мессенджер XChat, продвигавшийся Илоном Маском как сервис с шифрованием без рекламы. Несмотря на трафик из X.com, приложение исключили из сторов. Причиной мог стать отказ от трекинга, что мешает монетизации и антифрод-системам площадок. Кейс доказывает: отсутствие рекламных инструментов делает софт уязвимым, а медийная поддержка не спасает продукт, если его политика нарушает правила маркетплейсов.

➡️ Читайте на сайте: https://aff.top/blog/xchat-udalen-iz-app-store-i-google-play

🧠 Ещё больше инсайтов → в канале AFF.top

В Facebook Ads Manager появилась метрика Creative Diversity

Meta представила метрику Creative Diversity для оценки разнообразия креативов. Алгоритм Generative Recommender анализирует стиль, тему, тип хука и формат объявлений. Простое изменение цвета или ракурса теперь не работает: система пессимизирует похожие подходы. Для эффективного залива арбитражникам придется тестировать принципиально разный контент, чтобы алгоритмы лучше находили аудиторию. Инструмент выйдет из беты до конца года.

➡️ Читайте на сайте: https://aff.top/blog/v-facebook-ads-manager-poiavilas-metrika-creative-diversity

🧠 Ещё больше инсайтов → в канале AFF.top

Твои переписки с ChatGPT читают люди

OpenAI привлекает подрядчиков для ручного анализа диалогов с ChatGPT, чтобы повысить качество модели. В ходе проверки конфиденциальная информация пользователей попадает к третьим лицам. Для сферы CPA это несет прямые риски: уникальные связки, креативы и структуры лендингов, созданные нейросетью, перестают быть приватными. Главный вывод — любая информация, переданная ИИ, может быть изучена извне, что ведет к быстрому выгоранию профитных подходов …

➡️ Читайте на сайте: https://aff.top/blog/tvoi-perepiski-s-chatgpt-chitaiut-liudi

🧠 Ещё больше инсайтов → в канале AFF.top

The trust signal Google won't name: long clicks

The question: when the 2024 antitrust disclosures revealed Navboost, did they reveal a trust mechanism hiding in plain sight?

Arguably yes. Navboost, per testimony in the U.S. v. Google proceedings, aggregates click signals — and specifically the distinction between 'good clicks' and 'bad clicks,' where a long dwell with no pogo-stick back to results reads as satisfaction. This is not labeled a trust signal, but functionally it is one: users staying with content correlates with content that earned their confidence.

The supporting logic: trust is, behaviorally, willingness to keep relying on a source. Dwell, return visits, and branded re-searches operationalize that willingness without any need to parse credentials. A system that rewards satisfied long clicks is rewarding earned trust by proxy.

The counter-evidence is sharp. Click satisfaction is confounded by presentation, headline accuracy, and load speed — a page can earn long clicks by being merely adequate and well-formatted, not trustworthy. Engagement and trust correlate but are not the same construct, and conflating them is a known error.

Caveat: these signals are aggregated across many users over time and are heavily denoised. Individual or small-sample manipulation has historically been detectable and devalued; this is not a button you press.

What we still don't know: how Navboost-style signals interact with the explicit quality systems. The disclosures confirmed existence and rough function; they did not reveal how click data is reconciled against link and content signals when the two disagree.

Documenting everyday expertise where no formal credential exists

The question: your author has deep lived knowledge but no diploma in the subject — how do you make that legitimate expertise legible, given the QRG explicitly recognizes it?

The guidelines state that expertise can be everyday: someone who has lived with a condition, used a product for years, or practiced a craft can have valuable expertise even without formal credentials. The documentation playbook makes that visible:

— Establish the basis of experience concretely: duration, scope, and specifics. "Managed type 1 diabetes for twenty years" is documentable; "health enthusiast" is not.
— Surface the experience in the content itself, through the non-obvious details that only sustained practice produces.
— Provide verifiable traces where possible: a long track record of consistent contribution, community standing, original media from the lived experience.
— Be precise about the boundary. Everyday expertise on living with a condition does not extend to prescribing treatment; the QRG keeps that distinction, and crossing it undermines trust.
— Pair lived experience with professional review on the clinical claims, so each type of authority covers its own domain.

Caveat: everyday expertise is most clearly valued for experience-driven and community topics. For the strictest medical, financial, or legal claims, the QRG still expects formal professional expertise — lived experience complements it but does not replace it where stakes are highest.

What we still don't know: how algorithmic systems, as opposed to human raters, recognize everyday expertise, which carries far fewer machine-verifiable signals than a formal credential with a registry to check.

Google тестирует оплату за контент сайта

Google запустил тест программы AI Contribution Pilot, предлагая вебмастерам выплаты через Search Console за контент, повлиявший на ответы ИИ в Gemini и AI Overviews. Алгоритмы оценки вклада и формулы расчетов непрозрачны, а суммы пока скромные. Главный итог: создан важный прецедент прямой компенсации за использование авторских материалов в генеративной выдаче.

➡️ Читайте на сайте: https://aff.top/blog/google-testiruet-oplatu-za-kontent-saita

🧠 Ещё больше инсайтов → в канале AFF.top