{"slug": "claude-got-guidelines-now-what", "title": "Claude Got Guidelines, Now What?", "summary": "Anthropic published its 2026 Usage Policy update on October 8, adding a section titled \"Do Not Engage in Deceptive Campaigns or Artificial Activity\" that prohibits using Claude to manipulate the sources search engines or AI systems draw answers from, effective November 12. Anthropic said the section consolidates rules previously scattered across its elections, fraud, privacy, and disinformation sections and applies whether the deception is political or commercial, citing the example of a network of sites posing as unaffiliated sources that all corroborate the same claims. The author calls it the first time a frontier model provider has written \"don't game AI answers\" into its rules in plain language, drawing a parallel to Google's Webmaster Guidelines, which went unenforced for years before updates such as Florida in November 2003, Panda in February 2011, Penguin in April 2012, Medic in August 2018, and the March 2024 spam policies on expired domain abuse, scaled content abuse, and site reputation abuse.", "body_md": "On October 8, Anthropic published its [2026 Usage Policy](https://www.anthropic.com/news/2026-usage-policy-update) update, and most of the coverage will focus on weapons, surveillance, and elections. That makes sense, because those are the headline risks. But if you work in search, there is one bullet in the new [Usage Policy](https://www.anthropic.com/legal/aup) that you should print out and tape to your monitor.\n\nUnder a brand new section called “Do Not Engage in Deceptive Campaigns or Artificial Activity,” Anthropic now prohibits using Claude to manipulate the sources that search engines or AI systems draw answers from by seeding them with content that misrepresents its origin, authorship, or independence. The example they give is a network of sites posing as unaffiliated sources that all corroborate the same claims. If you’ve spent any time in the GEO corner of LinkedIn over the last two years, you know exactly which playbook that sentence describes.\n\nAnthropic says the new section consolidates rules that were previously scattered across its elections, fraud, privacy, and disinformation sections, and that it applies whether the deception is political or commercial. The updated policy takes effect November 12. As far as I can tell, this is the first time a frontier model provider has written “don’t game AI answers” into its rules in plain language.\n\nSo Claude got guidelines. Now what? To answer that, you have to look at the last company that wrote rules for how people compete for visibility, because we already know how that story played out.\n\n## We've Seen This Movie Before\n\nGoogle has always told you what it wanted. The original Webmaster Guidelines were explicit from the early 2000s: no hidden text, no cloaking, no doorway pages, no link schemes. The problem was never that the rules were unclear. The problem was that the rules were written years before Google could actually enforce them, and the industry did what any industry does with an unenforced rule. It treated the rule as a suggestion and the gap as an opportunity.\n\nThat gap is the whole story of SEO. The Florida update in November 2003 wiped out keyword-stuffed affiliate sites right before the holiday season. In 2006, Google removed BMW’s German site from the index for using doorway pages, which was a public execution designed to show that even big brands weren’t exempt. Content farms printed money for years until Panda arrived in February 2011 and hit roughly 12% of queries. Link schemes had been against the guidelines for nearly a decade when Penguin landed in April 2012 and turned years of purchased links into liabilities overnight.\n\nThe pattern kept repeating as the guidelines matured. The Quality Rater Guidelines spent years describing expertise and trust for “Your Money or Your Life” topics before the August 2018 “Medic” core update made health and finance sites feel it. Google’s Helpful Content system launched in 2022 to demote content made for search engines first. Then, in March 2024, Google [introduced three new spam policies](https://developers.google.com/search/blog/2024/03/core-update-spam-policies) for expired domain abuse, scaled content abuse, and site reputation abuse, alongside a core update that folded the helpful content system into core ranking.\n\n| When | What the guideline said | How enforcement arrived | \n|---|---|---|\n| March 2024 | Scaled content abuse, site reputation abuse, expired domain abuse | March 2024 core and spam updates, followed by manual actions | \n| August 2022 | Content should be made for people first | Helpful Content system, later absorbed into core ranking | \n| August 2018 | Expertise and trust for YMYL topics (Quality Rater Guidelines) | “Medic” core update | \n| April 2012 | No link schemes (in the guidelines since the early 2000s) | Penguin, made real-time in 2016 | \n| February 2011 | No thin or low-value content | Panda | \n| November 2003 | No keyword stuffing or hidden text | Florida update | \n\nWhile this is just a small subset of the many algorithm updates, notice the shape. The rule comes first. The tactic that breaks the rule scales anyway, because there is no immediate cost. Then a system arrives that makes the rule real, and everyone who built on the gap gets wiped out at once. The marketers who survived each cycle weren’t the ones who read the guidelines most closely. They were the ones who understood that the guidelines described where the algorithm was going.\n\n## Google Folded AI Search Into SEO, and That's the Problem\n\nGoogle is the one player that never wrote separate AI search guidelines, because it never admitted AI search was separate. Its [AI features documentation](https://developers.google.com/search/docs/appearance/ai-features) says there are no additional requirements or special optimizations needed for AI Overviews or AI Mode. The [May 2026 guide to optimizing for generative AI features](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) goes further, saying that from Google’s perspective optimizing for generative AI search is still SEO, and it waves off chunking, llms.txt, and rewriting content for AI systems as things you can ignore.\n\nI’ve already argued that [Google’s guidance on AI search is naive and self-serving](https://ipullrank.com/google-ai-search-guidance), and I stand by it. The [Content Warehouse leak](https://ipullrank.com/google-algo-leak) showed us how far Google’s public statements and its internal systems can drift apart, and the new guide flattens a reality where AI Mode, AI Overviews, and classic ranking are different systems run by different teams. Calling all of it “still SEO” isn’t a clarification. It’s the same move that turned mobile, voice, and AMP into more scope for SEO teams without more budget or authority, and it’s great for Google and lousy for the people doing the work.\n\nThe guide also ignores how these systems actually choose content. Retrieval computes similarity at the passage level, and synthesis runs comparisons between candidate passages before anything reaches the model, so chunking and writing for retrieval matter whether Google admits it or not. Compare that posture to Bing, which shipped [AI Performance in Bing Webmaster Tools](https://blogs.bing.com/webmaster/February-2026/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview) as an early step toward GEO tooling and has [explained publicly](https://blogs.bing.com/search/May-2026/Evolving-role-of-the-index-From-ranking-pages-to-supporting-answers) how its index is shifting from ranking pages to supporting answers. One company is telling you what’s happening. The other is telling you what it wants you to do.\n\nBut there’s a difference between Google’s recommendations and Google’s prohibitions, and the history above tells you which one to take seriously. The recommendations are positioning. The prohibitions are where enforcement lands. In the same guide, Google says that spinning up separate pages for every fan-out query primarily to manipulate rankings or AI responses violates its scaled content abuse policy, and that chasing inauthentic mentions won’t help because its AI features depend on the spam systems as much as the ranking systems. That isn’t advice about craft. It’s a line Google already has the machinery to enforce, courtesy of the March 2024 spam policies.\n\nSo read Google’s AI guidance the way you should have always read its guidelines. Discount the parts that tell you nothing has changed, and pay close attention to the parts that tell you what will get you hurt. That’s also the right lens for Anthropic’s update, because a usage policy is all prohibition and no reassurance.\n\n## What Anthropic's Policy Actually Says to Marketers\n\nBefore anyone overreacts, get the scope right. Anthropic’s [Usage Policy](https://www.anthropic.com/legal/aup) governs how people use Claude. It applies to anyone sending inputs to Anthropic’s products, including API customers and the end users of products built on Claude. It is not a webmaster guideline, and it says nothing about how Claude ranks, retrieves, or cites your content. Google’s guidelines tell you how to compete for visibility in its product. Anthropic’s tell you what you can’t do with its product.\n\nThat distinction matters less than you’d think, though, because most marketing teams now run on these models. If your content operation, your outreach, or your agents run on Claude, these rules apply to your workflow directly. Here’s what touches our world:\n\n- **Deceptive campaigns and artificial activity.** Fake personas, sockpuppets, astroturfing, fake reviews, and bots that claim to be human are prohibited. So is concealing the sponsorship of content meant to influence public opinion, distributing content through networks of ostensibly independent sites to hide a common source, and the AI search manipulation clause discussed above. Anthropic also bans building tools for any of this and selling it as a service, which should get the attention of anyone running a “citation seeding” offering.\n- **Spam.** Promoting or facilitating the generation or distribution of spam is prohibited under the fraudulent practices section.\n- **Other platforms’ guardrails.** Using Claude to circumvent the guardrails or terms of other platforms is prohibited. Think about what that means for automated Reddit seeding, mass account creation on review sites, or scraping that violates a site’s terms.\n- **Automated accounts.** Using automation for account creation or spammy behavior is listed under platform abuse.\n- **Attribution.** Submitting AI-assisted work without proper permission or attribution falls under the same section.\n- **Agents.** Anthropic states that users are responsible for making sure the agents they build or deploy comply with the policy, including the actions those agents take through tools, browsers, and connected systems.\n\nThe [announcement post](https://www.anthropic.com/news/2026-usage-policy-update) ties the new deceptive activity section to what Anthropic documented in its September threat intelligence report: state media outlets, propaganda offices, and commercial firms using Claude to run networks of fake accounts and fabricated news sites. Commercial firms are named right alongside governments. Anthropic isn’t drawing a distinction between a disinformation operation and a brand quietly building twenty “independent” review sites that all recommend the same product. Structurally, those are the same thing.\n\nThe enforcement mechanism is also worth noting. Anthropic says its Safeguards team uses detection and monitoring, and suspected violations can lead to warnings, throttling, suspension, or termination. Think of it as the manual action equivalent for your tooling. Losing access to the model your content operation depends on is a much faster penalty than waiting for a core update.\n\n## How OpenAI and Perplexity Got Here\n\nOpenAI and Perplexity have both built search products, but neither has published anything resembling Google’s spam policies. Their guidelines emerged from different pressures, and it shows.\n\nOpenAI’s guidance is about access, not quality. The [Publishers and Developers FAQ](https://help.openai.com/en/articles/12627856-publishers-and-developers-faq) says any public website can appear in ChatGPT search and tells you to make sure you aren’t blocking OAI-SearchBot. It explains that ChatGPT tags referrals with utm_source=chatgpt.com and that you should disallow GPTBot if you want to stay out of training. In December 2025, OpenAI [revised its crawler documentation](https://ppc.land/openai-revises-chatgpt-crawler-documentation-with-significant-policy-changes/) to separate search from training more cleanly and removed robots.txt compliance language for the user-initiated ChatGPT-User agent. On the manipulation side, OpenAI’s [Usage Policies](https://openai.com/policies/usage-policies/), last updated October 29, 2025, prohibit deceit, fraud, scams, spam, and impersonation in general terms, and its [ChatGPT agent policy](https://openai.com/policies/using-chatgpt-agent-in-line-with-our-policies/) specifically calls out generating false reviews or testimonials. What you won’t find is any language about seeding sources to influence ChatGPT’s answers.\n\nPerplexity’s guidelines were shaped by conflict. Its [crawler documentation](https://docs.perplexity.ai/docs/resources/perplexity-crawlers.md) recommends allowing PerplexityBot if you want to appear in results and publishes its IP ranges. The Perplexity-User agent, which fetches pages when a user asks a question, generally doesn’t follow robots.txt because Perplexity treats those requests as user-initiated. That position blew up in August 2025, when Cloudflare [accused Perplexity of stealth crawling](https://searchengineland.com/cloudflare-vs-perplexity-ai-crawling-460016) by switching user agents to get around blocks. Perplexity called the claims a misunderstanding of how AI assistants work. Its answer to the publisher relationship has been economic rather than editorial: a Publishers’ Program in 2024 that shares ad revenue, and Comet Plus in 2025, which shares subscription revenue based on human visits, citations, and agent actions.\n\nPut all four side by side and you can see how fragmented this is.\n\n| Platform | Where the guidance lives | What it says about manipulating AI answers | How it’s enforced | \n|---|---|---|---|\n|  | Search Essentials, spam policies, [generative AI guide](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) | Fan-out page spam is scaled content abuse; inauthentic mentions don’t help | Ranking systems, spam updates, manual actions | \n| Anthropic | [Usage Policy](https://www.anthropic.com/legal/aup) (effective Nov 12, 2026) | Explicitly bans using Claude to seed AI and search sources with content that misrepresents origin or independence | Account warnings, throttling, suspension, termination | \n| OpenAI | [Usage Policies](https://openai.com/policies/usage-policies/) ,[publisher FAQ](https://help.openai.com/en/articles/12627856-publishers-and-developers-faq) | General bans on deceit, spam, and fake reviews; nothing specific to answer manipulation | Account enforcement; crawler controls for publishers | \n| Perplexity | [Crawler docs](https://docs.perplexity.ai/docs/resources/perplexity-crawlers.md) , publisher revenue programs | No published quality or spam guidelines for sources | Not publicly documented | \n\nThere are two kinds of rules in that table. Google publishes rules about what it rewards and punishes in its results, and as I said above, only the punishment half deserves to be taken at face value. Anthropic and OpenAI write rules about what you’re allowed to do with their models. Nobody outside of Google has published the first kind for their AI search product yet, and that’s the gap you should be planning around.\n\n## The Protocol for a Post-Guideline World\n\nIf the Google era taught us anything, it’s that the time to adjust is when the rule is written, not when the enforcement lands. Here’s how I’d operate from here.\n\n### 1. Treat every policy as a preview of enforcement\n\nRight now, Anthropic’s rule governs the tool, not the results. But retrieval systems learn from the same patterns their makers are writing policy about. A network of sites corroborating the same claims is a detectable footprint, just as a link wheel was in 2011. Assume that what is banned on the input side today will be demoted on the output side tomorrow, and plan your program so that the day it happens is a non-event for you.\n\n### 2. Audit your AI search program for manufactured independence\n\nThe key word in Anthropic’s new clause is “independence.” Go through every tactic in your AI visibility program and ask whether it depends on making something look independent when it isn’t. Affiliate review networks you control, “best of” listicles you paid for without disclosure, coordinated Reddit and Quora answers from accounts that don’t say who they work for, and microsites built to echo your positioning all fail that test. If the tactic only works because the model can’t tell the source is you, it has an expiration date.\n\n### 3. Disclose sponsorship and authorship everywhere\n\nThe fix for most of the above is not to stop doing the work. It’s to stop hiding who did it. Paid placements can be labeled, contributed content can carry a real byline, and brand-owned comparison pages can say they’re brand-owned. A disclosed sponsored placement is still a mention, and it doesn’t put you on the wrong side of any of these policies.\n\n### 4. Use fan-out to find gaps and fulfill them with high information gain\n\nQuery fan-out is the most important concept in AI search right now, and it’s also the easiest one to abuse. The lazy move is to take every synthetic query a model generates and spin up a page for it, which Google explicitly calls scaled content abuse. The better move is to use fan-out analysis (it’s why we built Qforia) to see which subtopics the systems are reaching for that your content doesn’t cover, then fill those gaps with information the model can’t get anywhere else: original data, first-party experience, expert positions, and specifics that break from the consensus. A retrieval system pulling from twenty sources has no reason to cite a twenty-first page that says the same thing. Coverage gets you considered, but information gain is what gets you cited.\n\n### 5. Hold your agents to the same rules as your people\n\nAgentic content operations are where most teams will trip over these policies without meaning to. Anthropic says plainly that you’re responsible for what your agents do through tools, browsers, and connected systems. An agent that creates accounts, posts comments, or submits reviews at scale is your conduct, not the model’s. Write an internal policy for agent behavior that mirrors the strictest provider policy you use, and log what your agents actually do.\n\n### 6. Manage crawler access per platform, on purpose\n\nEach platform splits its crawlers differently, and your robots.txt is now a business decision. OpenAI separates OAI-SearchBot (search) from GPTBot (training). Perplexity separates PerplexityBot (indexing) from Perplexity-User (live fetches that generally ignore robots.txt). Google now offers a Search Console control for inclusion in its generative AI features. And when Google says it ignores llms.txt, remember that’s a statement about Google, not the ecosystem. If the systems you care about consume it, publish it. Decide what you want from each platform, whether that’s citations, training exposure, or neither, and configure each one deliberately instead of copying a blocklist from a forum.\n\n### 7. Invest meaningfully in the shared, owned, and earned media ecosystem\n\nAI search systems synthesize what the web says about you, which means your visibility is a function of your entire media footprint. Owned media gives the models something authoritative to retrieve. Earned media gives them corroboration that is actually independent. Shared media puts you into the conversations that fan-out queries pull from. “Meaningfully” is the operative word here, because most brands fund this with whatever is left after paid. Real earned media means digital PR, original research, and relationships with the publications and creators the models already trust, funded at the level you’d fund a channel rather than an experiment. This is the durable version of everything the manipulative playbook tries to fake, and it’s the only version that gets stronger when enforcement arrives.\n\n### 8. Measure inputs, channels, and performance separately\n\nWhen enforcement hits, you need to know which layer moved. Track your inputs (content coverage, mentions, crawl access), your channel metrics (citations and visibility per platform, treating Google’s Generative AI performance report in Search Console as one input rather than the source of truth), and your business performance. Teams that only track a single “AI visibility score” won’t be able to tell whether they lost ground because of a policy change, a model update, or a competitor.\n\n## The Gap Is Closing Faster This Time\n\nIt took Google nearly a decade to turn its link guidelines into Penguin. AI search won’t give you that kind of runway. Google has already folded AI search into the rules it spent twenty years enforcing, Anthropic has written AI answer manipulation into its policy by name, and the other platforms are one bad news cycle away from doing the same.\n\nThe guideline era isn’t the end of opportunity in AI search. It’s the beginning of the part where the work compounds instead of evaporating. Every cycle of SEO history rewarded the people who stopped hunting for the space between the rule and the algorithm and started building for where the algorithm was obviously headed. Claude got guidelines. Build like the enforcement is already here, because the next update is coming whether you’re ready or not.\n\nFor more on how retrieval, fan-out, and citation work across these platforms, see [The AI Search Manual](https://ipullrank.com/ai-search-manual).", "url": "https://wpnews.pro/news/claude-got-guidelines-now-what", "canonical_source": "https://ipullrank.com/claude-geo-guidelines", "published_at": "2026-10-12 17:30:39+00:00", "updated_at": "2026-10-12 17:50:34.780241+00:00", "lang": "en", "topics": ["ai-policy", "generative-engine-optimization", "ai-search", "artificial-intelligence"], "entities": ["Anthropic", "Claude", "Google", "LinkedIn"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/claude-got-guidelines-now-what", "markdown": "https://wpnews.pro/news/claude-got-guidelines-now-what.md", "text": "https://wpnews.pro/news/claude-got-guidelines-now-what.txt", "jsonld": "https://wpnews.pro/news/claude-got-guidelines-now-what.jsonld"}}