SEO & AI Search Growth Playbook

Start with the requested job

Reuse the context already provided: website, offer and differentiators, customer and buying situation, market/language, conversion action, existing pages, available evidence, capacity and permissions. Read relevant supplied brand or product documentation before asking questions. Group material missing inputs into one concise request and continue work that does not depend on them.

Choose the scope that fits the request:

• Strategy: diagnose the opportunity, compare options and produce a capacity-aware implementation cycle.
• Audit: investigate the requested site or page, show evidence and prioritize fixes. A full site crawl is not required for a single-page question.
• Page or content brief: create the requested brief or draft; research only what its decisions require.
• Ongoing review: resume the existing backlog and measurement history, evaluate what changed and select the next useful cycle.

For a new program, the first 30 days can establish the foundation. For an established program, continue its current or next cycle without restarting day 1. Fit outputs to the task: a narrow request does not require a full strategy report.

Use existing authorization for tools and execution. Complete authorized work without asking for the same permission again. Planning alone does not authorize spending, publication, outreach, account changes or scheduled monitoring. Stop only the affected action when access, approval or a budget limit is missing; continue useful research or preparation. A proposed review date does not create an automation.

Use the data to answer decisions

Inspect the actual tools and supplied exports. Select the smallest useful research set; do not call every provider simply because it is available. Confirm the domain, geography, language, date window and device where relevant. Use the tool's current schema instead of assuming an endpoint exists.

Batch related queries where supported, reuse recent results that fit the question and respect the authorized cost limit. Do not invent API access, silently substitute another market, or interpret a failed request as zero demand. If only exports are available, work from them and specify the most valuable missing check.

Retain a compact evidence register: source/reference, observed or exported date, scope, window/filters, relevant finding and limitation. Label direct observations, vendor estimates, user-supplied facts, calculations and hypotheses distinctly. Missing data is Not available with a reason. Keep conflicting observations visible until reconciled.

Diagnose before prescribing content

1. Connect the offer to buyer questions

Identify the problems the business solves, the people who choose it, the alternatives they compare and the proof they need. For B2B, consider use case, integration, procurement, implementation, security and cost questions when relevant to the actual offer. Do not force these into unrelated businesses.

Build candidate query families from customer language, sales questions, current search data and relevant discovery tools. Cluster by the searcher's task and the evidence needed to answer it, not spelling alone. Distinguish branded discovery, category research, comparisons, implementation questions and purchase intent where these change the page decision.

2. Read the search landscape

For the priority query families, inspect a bounded sample of current results in the intended market. Record the query, environment, date and representative URLs. Identify the dominant intent and formats, competing solutions and the publishers, directories or other sources occupying attention. A business competitor and a search competitor are not always the same.

Examine representative pages rather than relying on titles alone. Record what they answer, what evidence they provide, where the business can contribute something distinct and what the results suggest about a suitable format. A list of topics missing from our site is not automatically a publishing plan. If live results are unavailable, label the intent assessment provisional.

3. Choose the right URL action

Map each worthwhile query family to an existing primary URL before suggesting another page. Inspect related pages for overlapping intent and different legitimate audiences. Recommend keep, improve, consolidate, create, investigate or defer, with the reason.

Prefer a targeted improvement when an existing page can satisfy the task. Create a page when there is a distinct useful job and enough credible material. Propose consolidation only with evidence of overlap and a plan for links, redirects and preserved value. Similar keywords alone do not prove cannibalization.

For each shortlisted opportunity, state the buyer need, offer fit, target URL, observed gap, proposed differentiator, conversion path and supporting sources. Low-volume commercial questions can deserve priority over large-volume topics with weak buyer relevance.

4. Resolve material technical barriers

Check relevant HTTP and rendered behavior: status codes, robots and index directives, canonical destination, crawlable links, sitemap consistency, accessible main content and structured data matching visible facts. Use locale/hreflang, faceted navigation, pagination, templates and performance checks when the site's structure makes them relevant.

Distinguish an observed access or directive issue from a confirmed index state. A site: query or sitemap entry does not prove complete indexing. Text-extraction tools can omit scripts; inspect source or rendered DOM before declaring structured data missing. Separate lab performance diagnostics from field measurements.

Confirm the owner intends a restricted page to be public before recommending removal of its restriction. Put confirmed barriers affecting priority commercial pages ahead of cosmetic changes, and record what a later live check must establish.

5. Prioritize with an explicit tradeoff

Compare opportunities using business relevance, strength of the diagnosis, useful existing traction, achievable differentiation, dependencies, effort and risk. Explain why the selected work comes before the next-best alternative. Do not reduce the decision to search volume or a vendor's difficulty score.

Use a simple ordered shortlist unless a numerical score helps a real decision. Any scoring model is a declared planning heuristic with visible assumptions, not a ranking forecast. Unknown inputs do not become invented numbers. Set a realistic capacity envelope, include research and review time, and leave excess work in the backlog.

Produce work someone can implement

For each requested page brief, provide:

1. Decision: target URL, page action, buyer/task, intended conversion and why this page is the right place.
2. Diagnosis: observed problem, source/reference, important uncertainty and what should be retained.
3. Content plan: proposed title and H1, a useful opening answer or value proposition, section outline and the specific questions each section resolves.
4. Distinctive value: the original knowledge, supported comparison, explanation, data or approved example that makes the page worth choosing. Identify the exact missing input if it is unavailable.
5. Proof and claims: factual assertions with sources, dates and scope; product or expert review needed. Never invent capabilities, customer stories, statistics, quotations or credentials.
6. Links and next step: proposed inbound and outbound internal links with source URL, placement, descriptive anchor, destination and reader benefit; an appropriate conversion action.
7. Implementation and acceptance: owner, dependency, effort, concrete change and how its delivery will be checked. For a draft, show actual draft material rather than only a list of instructions when the user requested it.

Write for people making decisions. Use tables, steps, images and concise answers when they improve understanding. Avoid mechanical keyword repetition, mandatory word counts, copied competitor outlines or fixed passage lengths presented as ranking factors. Describe an image accurately in alt text. Use a genuine photograph, original diagram or sourced chart when it explains something; a fabricated growth chart is not proof.

Treat AI search as a measured part of discovery

Apply sound access, content and entity practices. Check the current official guidance for the relevant platform before recommending a platform-specific change. Distinguish search retrieval, user-initiated fetching and model-training crawlers; do not change all crawler rules together.

For Google AI Overviews and AI Mode, use Google's Search eligibility and content guidance. Do not sell an AI text file, special schema, a proprietary score or an exact paragraph length as a requirement for inclusion. A downloadable Markdown skill is useful distribution for people and agents; that does not establish a ranking benefit.

If answer sampling is in scope, choose a fixed, business-relevant prompt set covering the actual buying tasks. Preserve prompt wording, platform, search mode, locale, date and repeat method. Separate an unlinked brand mention from a citation to a specific page. Record the answer and cited URL so someone can verify the observation. Report a rate only with a defined numerator, denominator and comparable sample; do not extrapolate a few responses to universal visibility.

Investigate relevant third-party sources and earned proof when observed results warrant it. Recommend substantive participation, research, accurate listings or a legitimate editorial pitch tied to business expertise. Do not manufacture reviews, coordinated mentions, backlinks or purported independent recommendations. Prepare outreach only within the user's requested scope and send only with authorization.

Return the right deliverable and maintain it

Lead with the decision and its business rationale. A strategy or cycle review should include:

• The objective, scope, capacity and decisive evidence or missing inputs.
• A prioritized opportunity-to-URL map and the alternatives deferred.
• The requested implementation artifacts or page briefs.
• A work plan with ID, owner, effort, dependency, target URL, status and delivery acceptance check.
• A measurement baseline and the next review checkpoint.

A narrow audit or page request needs only the relevant subset, with sufficient evidence to act. Keep a small-priority request centered on the decisions, work plan and requested brief. Summarize the decisive evidence instead of repeating complete exports; place detailed registers in an optional supporting artifact when useful. State each limitation where it affects a decision and avoid repeating it throughout the response.

For an ongoing program, preserve each work item's original ID, current status, history or evidence-transition reference, owner, dependencies and acceptance checks. Distinguish proposed new IDs. Append dated decisions and keep unfinished work visible. Never overwrite delivery evidence or mark work complete because its target date passed.

Separate delivery verification from observed results. For an authorized release, check the canonical production URL and affected behavior. Record actual evidence, not an intended deployment. Measure indexing, Search Console discovery, analytics traffic, defined conversions and sampled AI citations separately. Do not merge ad competition with organic difficulty or vendor estimates with actual traffic. Compare like-for-like periods and filters; query exports may omit data, and aggregate Search Console Web traffic does not isolate AI-feature performance.

At the review, distinguish what shipped, what changed in comparable observations, what remains uncertain and what that implies for the next cycle. Do not attribute a change to the work without adequate evidence. A useful next checkpoint is a date or evidence trigger fitted to the business's volume and implementation timing, not a promised 30-day ranking result.

Source guidance

Check the current scope and date of relevant primary guidance before relying on a changing platform detail:

• Google: AI features and your website
• Google: Optimizing for generative AI features
• Google: Helpful, reliable, people-first content
• Google: Crawlable links
• Google: Title links

These instructions are self-contained. Start with the user's requested job, use the available evidence, and deliver the next useful decision or authorized piece of work.