What has changed about being found?
What has changed is that a buyer's question is now answered on several surfaces at once, and most of them synthesise an answer rather than listing pages. Google search still returns links; AI Overviews and AI Mode return an answer with sources; ChatGPT, Perplexity, Gemini, and Claude return an answer with citations; a voice assistant reads one answer aloud; video search surfaces chapters; and systems that fetch llms.txt orient from a Markdown map. The unit of success on most of these is a citation, and the unit of retrieval is a passage.
The temptation is to treat each surface as a separate discipline with its own tactics. This playbook takes the opposite view: the surfaces reward the same properties, because they are all trying to find the passage that answers a question directly, from a source they can trust, about an entity they can identify. Write for that and measure on each surface separately. Everything below is that one discipline, laid out in the order it should be applied, as it is applied on ftes.ai and by the Search Visibility Digital FTE.
What is the technical foundation?
The technical foundation has three parts. First, content and structured data must be present in the initial HTML, because many AI crawlers execute no JavaScript; on ftes.ai every page is statically rendered and every JSON-LD block is server-rendered inline, never injected by a client effect. Second, schema must be honest and complete: Organization on the root, Article or BlogPosting with author, publisher, and dates on content, FAQPage only where a visible FAQ exists, BreadcrumbList, Product or Service and ItemList for the catalogue, VideoObject only for real video, all generated from the same data that renders the page so they cannot drift.
Third, ordinary ranking health, because AI Overviews and AI Mode are built on core Search and cited pages observably rank for the query or a related one: indexable, fast, internally linked, one H1 per page, full metadata with canonical URLs, and a sitemap that includes everything. An llms.txt at the root, written as an honest map of the site's important pages, is a low-cost courtesy to systems that read it and not a ranking factor. The audit guide in the library turns these into nine checks with pass conditions.
What is the content discipline?
The content discipline has six rules, applied to every piece. Answer first: open with a 40 to 60 word direct answer that begins with a definition or verdict and could be lifted verbatim or spoken aloud. Chunk self-containment: every heading is a question or a claim, every section's first sentence restates the entity and topic, no section says as mentioned above, and sections run 120 to 220 words, because retrievers rank sections, not pages. Entity explicitness: one name per thing, used identically, with facts stated as entity-attribute-value triples and mechanisms named rather than benefits described.
Fan-out coverage: before writing, list the five to nine sub-queries an engine would split the primary question into, using the pattern what, how, cost, risk, versus, when, who, setup, and answer each in its own section or FAQ. Information gain: every piece carries at least one thing not already in the index, a named mechanism, a structural insight, a worked example, or a new comparison. Verifiability: no invented statistics, studies, customers, case studies, or quotes, ever; numbers are product mechanics, labelled arithmetic, or named public sources. Voice and video readiness follow from these: at least two FAQ answers under 30 words, and an H2 set that works as chapters.
Why do clusters matter more than pages?
Clusters matter more than pages because query fan-out assembles an answer from several sources covering different sub-questions, and a site whose pages consistently define and relate the same entities can be cited several times in one answer: a definition page for what, a pricing page for cost, a governance page for risk, a comparison page for versus. A single page, however good, is one candidate for one sub-query; a cluster is a candidate for all of them.
A cluster is hub-and-spoke: a definitional hub, spokes for each sub-question and role and industry and comparison, three related links on every piece, and consistent entity names and facts throughout. Inconsistency between pages undoes the cluster, because contradiction between a site's own pages lowers an engine's confidence in all of them. The ftes.ai library is organised this way, in clusters for concepts, hiring, roles, industries, economics, governance, comparisons, AI search, operations, and whitepapers, with every piece linked into its cluster and every mechanism named the same way in every cluster.
How is visibility measured honestly?
Visibility is measured by sampling, because no AI engine publishes a citation console. A fixed set of prompts representing the questions buyers ask, including the primary questions and their likely sub-queries, is run across each engine on a schedule, and for each prompt the record captures whether the site was cited, in what position, which competing sources were cited, and what the answer said, with the date. Results vary by session and location, so samples are repeated and reported as ranges. Search console data covers ordinary search where connected; referral traffic from AI surfaces and branded search lift corroborate the samples.
Honest reporting labels what is unmeasured as unmeasured. An engine not sampled, a metric that cannot be computed from the data, a voice surface not tested: each is stated rather than estimated. On ftes.ai this is a design rule for the performance view, and it is the rule the Search Visibility FTE reports under. A visibility report that shows only improvements is a marketing document; one that shows the misses and the gaps is the one a marketing leader can act on and defend.
How does a Search Visibility FTE run the playbook?
A Search Visibility Digital FTE runs the playbook as a loop. It audits the technical foundation and reports fails. It measures the prompt set and records citations. From the gaps, it proposes topics with a worth-it verdict: whether the topic is worth pursuing given the organization's own data, stated as a verdict with reasons rather than a list of keywords. For approved topics, it drafts pages to the content discipline, passes each through the fact gate, which blocks any claim without a source, and submits them to Approvals with the sources attached.
The manager approves, edits, or sends back with a reason; approved pages publish through the CMS the organization uses, or through a pull request to its site repository, or through the publishing endpoint a customer installs in their own Next.js site. Then the FTE measures again and reports, in the first person, what moved and what did not. Every step is in the audit trail, every claim is sourced, and nothing publishes without a named person. The FTE does not promise rankings; it runs the discipline and reports the record.
What does the playbook refuse to do?
The playbook refuses tactics with no documented or observable basis: hidden text for AI, schema for content not on the page, magic word counts, keyword stuffing, manufactured mentions, and claims that a specific trick guarantees citation. Several of these violate engine guidelines and risk ordinary ranking; all of them are superstition dressed as strategy. It also refuses to report precise citation rates for tactics without a sampling method, and it refuses to estimate what it cannot measure.
Most of all, it refuses invented claims. An invented statistic is the fastest way to lose on every surface at once, because generative engines cross-check sources and an uncorroborated figure marks the page and the domain as unreliable. Honesty is not a constraint on the playbook; it is the playbook's central mechanism, because the passages that win are the ones an engine can verify, and the only reliably verifiable claims are true ones about mechanisms you own and facts the public can check.
What should a team do first?
A team should first run the nine-check audit and fix the technical fails, which are fast. Second, record a measurement baseline: the prompt set, sampled once across engines, dated. Third, list the primary questions the business wants to be cited for and their fan-out sub-queries, and map which pages answer which; the gaps are the content plan. Fourth, rewrite the pages that already rank for target sub-queries to the content discipline, because they are closest to citation. Fifth, build out the cluster from the gaps, with information gain in every piece and the fact gate on every claim.
Then measure on the schedule and report honestly. The loop compounds: each measured cycle sharpens the prompt set, each cluster addition strengthens the entity, and each honest report builds the case for the next investment. The rest of the ftes.ai AI-search cluster covers each step in depth, from measuring citations to structuring for retrieval to voice and video readiness, and the Search Visibility FTE is built to run the whole loop as a managed direct report.
How do you apply the playbook in 90 days?
Apply the playbook in 90 days in three phases. Days one to thirty: run the nine-check audit and fix the technical fails, which are fast; write the prompt set covering the primary questions and their fan-out sub-queries; sample it once across every engine that matters and record the baseline, dated; and unify the entity names across the site and the listings you control. Days thirty-one to sixty: rewrite the pages that already rank for target sub-queries to the content discipline, because they are closest to citation, with every claim passing the fact gate and every section standing alone; publish through the gated route; and keep sampling weekly.
Days sixty-one to ninety: build out the cluster from the uncovered sub-queries, one piece per gap with information gain in each, linked three ways into the cluster; add voice-ready FAQs and, where a real video exists, chapters and a transcript; and hold the first quarterly comparison against the baseline, reporting citation presence per engine with sample sizes, console changes, referral and branded trends, and the engines and metrics still unmeasured. The Search Visibility FTE runs all three phases as its function, with its manager approving each page and reading the honest report at the end.
What objections does the playbook meet, and what are the honest answers?
The first objection is that AI search cannot be measured, so the playbook cannot be judged. The honest answer is that it cannot be measured from a console, and it can be measured by sampling: a fixed prompt set, run on a schedule, with citations recorded and dated, corroborated by referral and branded trends, and reported with what remains unmeasured. The second is that engines change too fast for any discipline to hold. The honest answer is that the discipline is derived from what engines are trying to do, find a direct, self-contained, trustworthy passage about a clear entity, which has not changed as the engines have, and that the sampling record is how you notice if it does.
The third is that this is just SEO with new words. The honest answer is that the foundations are shared and the section-level discipline is new, and that a page can rank in SEO and be invisible to citation if its answer is buried or unverifiable. The fourth is that a vendor promised a guaranteed citation tactic. The honest answer is that no such tactic is documented, that several claimed ones violate engine guidelines, and that the playbook's refusal to invent claims is not a limitation but the mechanism by which its pages become safe citations. Each answer names a method rather than asserting a result.