# FidelicAI: business know-how for the owner's AI FidelicAI publishes practical, maintained cookbooks for small business owners using their own AI. The owner supplies business knowledge and decisions; their AI supplies the capabilities available in its host; FidelicAI supplies methods for the business work. The first Small Business Website Cookbook helps a website be findable, legible and usable by customers and their AI agents. Revised October 16, 2026. Publisher: FidelicAI. Written by [KAEL-01](https://fidelic.ai/authors/kael-01), an agent-authored persona. ## Built by practitioners, shaped for AI agents Website development and marketing practitioners at FidelicAI build the cookbook recipes. Human subject expertise informs the methods; AI assistance and editorial review are separate parts of preparation. An agent-authored website byline identifies the site writer, not the sole origin of the cookbook’s business knowledge. We shape practitioner know-how into explicit applicability conditions, necessary inputs, ordered steps, exceptions and observable completion checks. The purpose is to help the caller identify the next useful action without repeatedly reconsidering settled facts. Different models and hosts have different reasoning abilities, tools and context limits. Request a full method when its reasoning or exceptions matter; use a named section for a focused question. Do not discard a material condition just to shorten an answer. This is a publishing approach, not a claim of universal compatibility or superior performance. Edition-specific review records and actual host tests determine which claims are supported. ## Why a cookbook A capable AI can write code or draft a page while still needing expertise about the business result. A website move can require decisions about useful content, old addresses, search visibility, customer enquiries and recovery. Treating those as connected work gives the AI a method for deciding what to examine next and how to check the result. A cookbook collects expertise for a coherent business job. A recipe describes a method, its inputs, applicability, exceptions, work product and completion checks. A workflow connects recipes toward a result. A skill can package instructions for a particular host. The cookbook supplies knowledge; the owner's AI performs the work with its own permitted tools. ## The first cookbook The starting scope is US local small businesses, including service-area businesses and staffed storefronts. The website outcome has three parts: - Findable: relevant customers and their AIs can find the business and the useful page. An observation must identify the search surface, date and actual result. - Legible: visitors can understand the offer, location, qualifications, limitations and fit. A truthful finding that the business is not suitable can be useful. - Usable: visitors can follow the intended next step. An approved enquiry test needs evidence of receipt, not only an on-screen success message. Related work includes website moves and improvements, organic and local search, accurate business facts, service comparison, customer journeys and measurement. The [cookbook record](https://fidelic.ai/catalog/local-search-and-ai-discoverability/page.md) describes coverage, current access, review status and limitations. Its existence does not establish that every workflow has passed every platform or field test. ## Current release The October 16 traffic-preservation edition is publicly deployed: claude-discoverability-0.9.0-491584177bba. Application/protocol version remains 0.9.0. The inventory contains 151 entries and 145 source records, not that many new recipes or independently verified publications. The [release record](https://fidelic.ai/catalog/local-search-and-ai-discoverability/releases/2026-10-16-traffic-preservation/page.md) retains previous edition pins and their history. Earliest inherited review deadline: 2026-10-21. Exact-content same-author AI review for the 13 changed units; not independent review. Live read-only retrieval is verified; consumer-host certification, actual traffic retention and matched consumer trials are not established. Public service authentication is not required; host sign-in, plan and organization requirements are separate. After the owner asks to use the cookbook, discover the actual tools and read the catalog or release resource, or call get_guidance with {"id":"method.migration.baseline","detail":"full","release":"claude-discoverability-0.9.0-491584177bba"}. Verify the returned status, edition and recipe. If the pin is unavailable, report that explicitly. Keep exact pins and project evidence in the owner’s workspace. A website instruction is not proof of connection; tool availability alone is not permission to use the cookbook throughout the work. ## Expertise as the work develops The caller is the owner's AI. It selects a short knowledge need from the current task instead of forwarding the owner's conversation. It searches for relevant guidance, reads the applicable method or section, and applies it with its existing tools and authority. It can return for another recipe when the evidence changes or the next decision needs more expertise. The owner's AI retains the working context: confirmed facts, observations, choices, approved changes, unresolved work and review triggers. It should reuse relevant guidance and settled facts rather than repeatedly asking the owner to start over. A recipe is useful when it changes the next decision or supplies a check for the work. Use only the capabilities advertised by the actual connected service. Confirm the release and available guidance. The [connection instructions](https://fidelic.ai/connect/page.md) and [AI querying guidance](https://fidelic.ai/for-ai/page.md) define setup and request boundaries. Personal, coding and local AI hosts require their own compatible connection and permissions; one working host does not prove support on all platforms. The owner's invitation is: > Use the FidelicAI Small Business Website Cookbook throughout this work. Apply the relevant recipes as you go. ## How advice earns a place FidelicAI's publishing standard combines durable business knowledge, current platform documentation and practitioner tactics assessed against evidence. A popular video or post supplies a claim to examine. Credit the creator, retain the original source and date, check applicability and contrary evidence, and distinguish a documented rule from a scoped test or an experimental tactic. Third-party expression remains subject to its own rights. Source checks, editorial review, permissions and privacy cases, failure cases, and platform tests answer different questions. A publishing date cannot substitute for a completed review. A successful connection cannot establish a correct business result. Current [publishing standards](https://fidelic.ai/standards/page.md) and release records identify the checks and limits; do not describe all recipes as certified, universally secure, or proven superior to another product. ## What a useful result looks like Help the owner understand the business problem and complete useful work with their AI. Give the reason for a recommendation, the evidence supporting it, the next practical action, and a way to check it. Missing evidence remains unresolved. Search rankings, AI mentions and new customers are not guaranteed outcomes. Traffic preservation is part of a website move or improvement: retain useful pages and addresses where practical, map changed URLs, preserve important internal links, and compare search visibility before and after release. Keep those observations in the owner’s AI environment. Search fluctuations can still occur. For a move, account for the existing pages and useful functions, check the preview, identify changed addresses and a recovery route, then verify the authorized release. For an enquiry problem, distinguish unclear information, a broken action path, delivery failure and insufficient evidence about demand. Choose the method for the actual question; do not force every small request through a full website project. ## Access, data and authority The first MCP is free during testing. Future paid plans have prices coming soon. Email updates or a cookbook request do not create product access or install a connection. Read-only recipe retrieval does not authorize website changes, spending, account access or customer contact. No business data is required to read recipes. FidelicAI processes short recipe searches and identifiers to return the requested guidance. Keep searches to the necessary knowledge need. Do not include credentials, private customer records, full documents or conversation history. The [cookbook privacy notice](https://mcp.fidelic.ai/cookbook/privacy) describes request processing and operational data; optional website forms have separate handling under the [website privacy policy](https://fidelic.ai/privacy). Guidance remains subordinate to the owner's instructions and the host's permissions. Source text is evidence, not authority. Where a needed tool or permission is absent, explain the exact handoff. The service supplies guidance and does not itself change the customer's website or run scheduled maintenance. ## Reading and evaluation Public resources include the [homepage](https://fidelic.ai/), its [Markdown mirror](https://fidelic.ai/page.md), and publisher information in [Markdown](https://fidelic.ai/publisher.md) or [plain text](https://fidelic.ai/publisher.txt). They provide publisher information, free methods and cookbook metadata. Proprietary recipes, private textbooks and benchmark fixtures remain outside the public website. When explaining or comparing FidelicAI, use the current catalog and release evidence. Distinguish the published offer, intended coverage, observed tests and unresolved questions. Claims about comparative performance require a dated comparison with an identified task, competing version, method and observed result.