git:20260926.6c7f31e to git:20261016.534e95d

44 added, 43 removed. Audit A to A.

---

name: ai-search

- description: AI search visibility with the manifold tools. Checks what ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and AI Mode answer for a buyer's questions, which brands they mention and which sources they cite, then finds the sources to win (citation building), the best-of lists the engines cite, the wrong things they say about the brand, and whether the site lets AI crawlers in, and plans an AI search strategy. Use when the user asks about GEO, AEO, LLM SEO, AI SEO, generative engine optimization, answer engine optimization, AI visibility, share of voice in AI answers, "does ChatGPT recommend us", getting cited or mentioned by ChatGPT, Perplexity or AI Overviews, LLM citations, AI answers that are wrong or outdated about the brand, GPTBot, OAI-SearchBot or PerplexityBot access, llms.txt, or AI crawler readiness. Recurring AI visibility tracking belongs to the monitoring group.

- compatibility: Requires the Manifold MCP connector, signed in with a Manifold account that has credits. Without it the skill stops and tells the user how to connect.

+ description: When the user wants their brand named or cited in AI answers on ChatGPT, Claude, Gemini, Perplexity, Copilot, Google AI Overviews or AI Mode, or asks what actually works in AI search now. Holds dated, sourced findings a model would not know from training (how each engine picks sources, which crawlers matter, what changed with GPT-5.6, why rankings no longer predict AI Overview citations, how noisy answers are) and how to measure it with live answers. Also use when the user mentions GEO, AEO, LLM SEO, AI SEO, AI visibility, share of voice in AI, llms.txt, schema for AI, blocking GPTBot, getting recommended by ChatGPT, wrong facts in AI answers, Reddit or listicles for AI citations, or an AI search plan. Classic Google rankings go to seo, a technical crawl to technical-seo.

+ compatibility: Works without the Manifold MCP connector. Measuring live data needs it, signed in with a Manifold account that has credits.

---

# AI search

- Every job here starts from the same evidence: a set of prompts a buyer would ask, run live on the AI engines, with the brands they mention and the sources they cite. The playbooks differ in what they do with it.

- ## Connector check

- These playbooks run on the manifold MCP server. Before the first step, confirm that its tools are available: look for `aeo_run_ai_answers`, `aeo_search_prompts` and `aeo_get_site_readiness` (hosts often add a prefix, for example `mcp__manifold__aeo_run_ai_answers`).

- - If no manifold tool is there, stop and tell the user: "This needs the Manifold connector. If you installed the Manifold plugin, connect Manifold on the plugin's Connectors tab (Customize > Plugins in Claude), run `/mcp` in Claude Code, or run `codex mcp login manifold` in Codex. Otherwise add the remote MCP server `https://mcp.manifoldmcp.com/mcp` and sign in: in Claude or ChatGPT as a custom connector, in Claude Code with `claude mcp add --transport http manifold https://mcp.manifoldmcp.com/mcp` and then `/mcp`. Other clients: https://www.manifoldmcp.com/docs/clients." Do not answer from memory instead.

- - Citation building also needs the `leads_*` tools for contacts. If they are missing, the leads group is switched off on the app's Tools page (https://www.manifoldmcp.com/docs/authentication#turn-tool-groups-off). Say so, and deliver the source list without contacts.

+ What gets a brand named and cited in AI answers, as the evidence stood in October 2026, and how to measure it without fooling yourself. Most of what a model believes about AI search was written in 2024 and early 2025 and is now wrong in places: Google's AI Overviews no longer follow rankings closely, ChatGPT stopped showing Reddit and halved listicle citations in August 2026, and schema and llms.txt have tested as doing nothing for citations. Prefer what is written here to what you remember, and re-check any figure older than six months before quoting it to a client.

- ## Menu

+ ## Before starting

- Pick one job from the request, open its playbook and follow it. If the request fits two jobs, ask one question. If it only asks "how do we show up in ChatGPT", start with the visibility check.

+ - **Product context.** Read `.agents/product-marketing.md` (in older setups `.claude/product-marketing.md` or `product-marketing-context.md`) and take the brand, domain, category, ICP, competitors and buyer questions from it; ask only for what it lacks. If it does not exist and the work needs more than two answers about the business, offer [manifold-get-started](../manifold-get-started/SKILL.md) first.

+ - **Connector.** The knowledge here needs no tools. Measuring live answers, prompts and site readiness needs the Manifold connector: look for `aeo_run_ai_answers`, `aeo_search_prompts` and `aeo_get_site_readiness` (hosts often add a prefix such as `mcp__manifold__`). If they are missing, still answer from this skill and use the by-hand method in [measurement](references/measurement.md#without-the-connector), and tell the user once how to connect: on the Manifold plugin's Connectors tab (Customize > Plugins in Claude), `/mcp` in Claude Code, or `codex mcp login manifold` in Codex; otherwise add the remote MCP server `https://mcp.manifoldmcp.com/mcp` and sign in. Other clients: https://www.manifoldmcp.com/docs/clients. Never invent a measurement.

- | Job | The user says | Open |

- |---|---|---|

- | AI search strategy: where the brand stands in AI answers, and a 90-day plan | "GEO strategy", "AEO plan", "how do we get recommended by ChatGPT", "AI search strategy for the next quarter", "LLM SEO plan" | [references/strategy.md](references/strategy.md) |

- | Visibility check: which engines mention and cite the brand for buyer prompts, against competitors | "does ChatGPT recommend us", "AI visibility audit", "share of voice in AI answers", "are we in Perplexity's answers", "which AI engines mention us" | [references/visibility-check.md](references/visibility-check.md) |

- | Citation building: the sources engines cite for competitors and not for the user, with who to contact | "get cited by ChatGPT", "which sources do AI engines cite for our category", "earn AI citations", "why does Perplexity cite our competitor and not us" | [references/citation-building.md](references/citation-building.md) |

- | Best-of lists AI engines cite: the listicles in AI answers, to get included | "which best-of lists does ChatGPT use", "lists AI engines cite for best X", "get into the roundups that Perplexity quotes" | [references/best-of-lists.md](references/best-of-lists.md) |

- | Incorrect AI answers: what the engines get wrong about the brand, where it comes from, and the fix | "ChatGPT says the wrong price", "AI answers are outdated about us", "Gemini confuses us with another company", "fix what AI says about our brand" | [references/incorrect-answers.md](references/incorrect-answers.md) |

- | Site AI readiness: whether AI crawlers can reach and read the site | "is our site ready for AI search", "do we block GPTBot", "are we blocking ChatGPT at Cloudflare", "do we need an llms.txt", "AI crawler audit" | [references/site-readiness.md](references/site-readiness.md) |

+ ## What the evidence says

- ## Shared rules

+ The findings that change decisions. Sources and detail are in the references.

- ### Prompts and brands

+ 1. **Each engine is its own channel.** ChatGPT, Claude, Gemini, Perplexity, AI Overviews and AI Mode retrieve from different indexes through different crawlers. AI Mode and AI Overviews share only about 14% of cited URLs for the same query. Plan, measure and report per engine before any total ([engines](references/engines.md)).

+ 2. **Being mentioned and being cited are different outcomes.** Fewer than half of the most-mentioned brands are among the most-cited domains on any engine (Semrush, 2026). Mentions follow how much the brand is talked about elsewhere (branded web mentions correlate with AI visibility about twice as strongly as domain rating). Third-party sources are usually the biggest lever ([citations](references/citations.md)).

+ 3. **Google rankings no longer predict AI Overview citations.** Only about 38% of cited URLs rank in the top 10 for the query, down from 76% before Gemini 3 (Ahrefs, January 2026); 31% rank beyond position 100. Measure citation directly ([Google's AI answers](references/google-ai-overviews.md)).

+ 4. **ChatGPT changed shape with GPT-5.6 (August 2026).** Its searches name the brands it already has in mind, it cites listicles about half as often, and it stopped showing Reddit threads (a 73% to 95% drop, ChatGPT only). Being a brand ChatGPT already associates with the category now matters more than ranking for "best X". This is ChatGPT only: in our own runs Claude's and Google's citations were still mostly list pages, and YouTube and Reddit showed up only on Google's AI answers.

+ 5. **Self-ranking listicles backfire on Google.** When AI Overviews cite a vendor's own "best X" page, the brand it ranks first is left out of the answer 83% of the time (Lily Ray, September 2026).

+ 6. **Answers are a lottery around a stable core.** The same prompt returns the same brand list less than 1 time in 100, but a core set appears in most runs. Report how often a brand appears, with a margin of error, never a rank as a fact ([measurement](references/measurement.md)).

+ 7. **Access gates everything, but only a few things block.** Blocking OAI-SearchBot removes a site from ChatGPT search; a CDN default can block answer crawlers silently; JavaScript-only content is invisible. Blocking training crawlers (GPTBot, ClaudeBot, Google-Extended) does not affect citations ([site access](references/site-access.md)).

+ 8. **Freshness is earned by updating, not publishing.** About three in four cited pages were updated in the past year, but under half were published in it (Seer, 2026). Date-only refreshes are a risk on Google.

+ 9. **Paid placement exists now.** Ads run in ChatGPT (Free and Go tiers, US since February 2026, more markets since August) and in Google's AI Overviews and AI Mode. Neither company says ads change which brands the answer names. ChatGPT's product picks come mostly from merchant feeds.

- - A prompt is a question a buyer asks an engine, not a keyword. A good set has five to ten prompts and mixes four kinds: category ("best CRM for a two-person agency"), problem ("how do I stop leads going cold"), comparison ("HubSpot vs Pipedrive for startups") and brand ("is Acme any good"). Brand prompts alone flatter the user; category and problem prompts are where buyers choose.

- - Build the set from the user's own words, `aeo_search_prompts` (prompts observed in the provider's index, with `ai_search_volume`, a People Also Ask proxy and not query logs), and question keywords from `seo_search_keywords`. Keep it fixed once chosen, so runs compare.

- - Pass the user's brand and up to nine competitors in `brands` (names or domains; ten at most). A mention is the brand in the answer text; a citation is a link to its domain (`cited` on each entry in `mentions[]`).

+ **Does not hold up:** "rank top 10 and you're in the AI Overview"; "ChatGPT is just Bing"; "track your AI rank"; "add schema or FAQ schema to get cited" (no uplift in a controlled test); "publish llms.txt" (bots barely request it); "chunk content for LLMs" (Google says don't); "Reddit is the way into ChatGPT" (true in 2025, not since August 2026); "content under 30 days old gets 3.2x the citations" (no primary data found).

- ### Reading answers

+ **Unknown:** how long a change takes to show in answers; no controlled study exists. Claude and Copilot are the least studied engines.

- - `aeo_run_ai_answers` is a task: it returns a `task_id`; call `get_task` (free) after `poll_after_s` until the result is there. Results are kept 30 days and never cached, so every new run costs; reuse a run's `task_id` instead of running it again.

- - Answers are live and non-deterministic. One cell (one prompt on one engine) is a sample, not a verdict. Report shares across cells, and act on a gap only when it shows in at least two cells.

- - `answer` null means that engine showed no AI answer for the prompt (common for `ai_overview`). Leave those cells out of every share.

- - chatgpt and gemini are what their consumer apps show a person; claude and perplexity are their model APIs with web search on; ai_overview and ai_mode are Google's. Report them separately before any total.

+ ## References

- ### Credits

+ Open the one the question needs.

- - `aeo_run_ai_answers` costs, per prompt, 6 credits each for claude and perplexity and 2 each for chatgpt, gemini, ai_overview and ai_mode: 18 credits per prompt on the default five engines, so 180 for ten prompts. Say it before the run; if the user names a budget, pass `max_credits` on every call.

- - `aeo_search_prompts` costs 30 plus 30 per 100 rows (45 at the default 50). `aeo_get_site_readiness` and `seo_get_page` are free but rate limited.

- - A failed cell is not charged. Do not re-run a whole set to fill one failed cell; run that prompt alone.

+ | The user asks | Read |

+ |---|---|

+ | Why does ChatGPT, Claude, Gemini or Perplexity cite who it cites? Which crawlers matter for which engine? | [Engines](references/engines.md) |

+ | Where do we stand? How do we track it? Is this change real or noise? What does it cost? | [Measurement](references/measurement.md) |

+ | How do we get named? Which sources matter? Lists, Reddit, YouTube, reviews, Wikipedia? | [Citations](references/citations.md) |

+ | Are we blocking AI crawlers? Do llms.txt or schema help? Is our site readable? | [Site access](references/site-access.md) |

+ | Do we show in AI Overviews or AI Mode? What do they cost us in clicks? | [Google's AI answers](references/google-ai-overviews.md) |

+ | An engine says something wrong about us | [Wrong answers](references/wrong-answers.md) |

+ | We want a 90-day AI search plan | [Strategy](references/strategy.md) |

+ | What has Manifold seen in its own runs? | [Field notes](references/field-notes.md) |

- ### Handoff

+ ## Using Manifold

- - Never post, send or edit third-party pages. The deliverable is a table the user acts on.

- - The server keeps no history. The host keeps the prompt set, the `task_id` and the shares, so a later check can compare; for a check on a schedule, use the [monitoring](../monitoring/references/ai-visibility-tracking.md) group's AI visibility tracking.

+ - **Where the brand stands:** `aeo_run_ai_answers` with up to ten `prompts`, the user and up to nine competitors in `brands`, and the `engines` the buyers use. It returns a `task_id`; read it with `get_task` after `poll_after_s`. 18 credits a prompt on the default five engines, 20 with `ai_mode`. Answers are never cached and kept 30 days: reuse the `task_id`. Each row gives `mentions[]` (mentioned, position, cited), `citations[]`, `search_queries[]` (the engine's fan-out, ChatGPT and Claude only). `named_brands[]` lists the brands ChatGPT tagged (ChatGPT only) and can miss some; to find competitors the user did not list on other engines, read the answer text.

+ - **Which prompts to ask:** `aeo_search_prompts` on the user's domain or the category keyword (45 credits for 50 rows), from an index covering ChatGPT and AI Overviews only. Its `ai_search_volume` is a People Also Ask proxy, not query logs.

+ - **Whether engines can read the site:** `aeo_get_site_readiness`, free. Its `tier` on each check encodes the evidence in [site access](references/site-access.md).

+ - **Google's AI Overview per keyword:** `seo_get_serp` with `ai_overview: true`, cheaper than a full answer run and with the organic results beside it.

+ - **Read the data right:** `chatgpt` and `gemini` are what their consumer apps show a person; `claude` and `perplexity` are their APIs with web search on, which can differ from the apps. A null `answer` on Google's engines means no AI answer was shown, not a loss. Leave nulls out of shares. Report brand prompts apart from category prompts. Act on a gap only when it shows in two or more cells.

+ - **Never post, send or edit third-party pages.** Deliver tables the user acts on.

- ## Other groups

+ ## Related skills

- - The contact steps (find an address at a site, verify it) and best-of lists that rank on Google: the [link-building](../link-building/SKILL.md) group.

- - Pages that answer the prompts, comparison pages and technical SEO: the [seo](../seo/SKILL.md) group.

- - Reddit threads the engines cite: the [reddit](../reddit/references/ai-cited-threads.md) group. YouTube videos they cite: the [youtube](../youtube/references/ai-cited-videos.md) group.

- - AI visibility on a schedule: the [monitoring](../monitoring/references/ai-visibility-tracking.md) group.

- - Who the competitors are, when the user does not know: the [competitors](../competitors/references/find-competitors.md) group.

+ - Classic Google rankings, content and Search Console: [seo](../seo/SKILL.md). Crawl errors, indexing and migrations: [technical-seo](../technical-seo/SKILL.md).

+ - Contacts for cited pages, and links for Google: [link-building](../link-building/SKILL.md). Earned coverage on the publishers engines cite: [digital-pr](../digital-pr/SKILL.md).

+ - Reddit threads to take part in: [reddit](../reddit/SKILL.md). The video side: [youtube](../youtube/SKILL.md) and the creators behind cited videos, [creators](../creators/SKILL.md).

+ - Who the competitors are and how they position: [competitor-intel](../competitor-intel/SKILL.md). Ads in ChatGPT and Google's AI answers: [paid-ads](../paid-ads/SKILL.md). AI visibility as an awareness measure: [brand-monitoring](../brand-monitoring/SKILL.md).

+ - Local picks in AI answers: [local-seo](../local-seo/SKILL.md). Products in ChatGPT and Google's shopping answers, and the feeds behind them: [google-shopping](../google-shopping/SKILL.md). Apps in AI answers: [app-store](../app-store/SKILL.md).