SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsai-generative-search-optimisation

ai-generative-search-optimisation

Design & creation

View repository

Updated 10-16 03:40

About this skill

AI Generative Search Optimisation (GEO)

Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com, +256 784 464178.

Plans evidence-bounded discoverability across AI answers, search, social profiles, communities and the destination a customer uses next, for the strategist and client reviewer. It is a planning and audit route, not a promise of inclusion, ranking or sales.

Use When

  • We never show up when customers ask ChatGPT, Perplexity, Gemini or Google's AI answers about our category.
  • The client talks about SEO, AEO, GEO or AIO and wants a programme-level plan that separates what is supportable from hype.
  • Tie social profiles, native content, reviews and off-site reputation to the website and the action customers take next.
  • Set up a measurement loop that keeps prompt observations, referrals, self-reported source and qualified enquiries apart.
  • Check that the business name, offer and facts are consistent across owned and third-party sources so AI tools describe us correctly.

Do Not Use When

  • seo-geo-optimisation for one page or article.
  • ai-readiness-diagnostic for a general AI maturity, data or team assessment.
  • paid-search-advertising for paid Google Ads visibility.
  • Stop before promising inclusion, ranking or citations, changing robots or firewall rules, or publishing without authority; record unverified claims as NOT_ASSESSED.

Required Inputs

Artefact Source/provider Required? If absent
Business name, offer, audience, market, goal, channels and the decision the plan serves Approved brief and owner-approved client fact sheet Yes Stop the affected recommendation; state a narrow assumption only where safe.
Existing profiles, content, canonical destinations and customer questions Supplied URLs, exports, comments/DMs, CRM or platform evidence Yes Audit only what was supplied; mark profile and entity consistency not assessed for the rest.
Analytics, referral data, "How did you find us?" answers, prompt observations and (where authorised) logs Platform exports, Search Console, Bing AI performance report, client Conditional Mark the check not assessed; do not infer visibility or performance.
Current platform, market, legal, privacy, rights and AI-search claims Social source register and Digital Research verification Yes for material claims Quarantine the claim as NOT_ASSESSED and narrow the deliverable.
Approval, access, budget, language, accessibility and moderation constraints Accountable owner Conditional Stop publication, spend, collection, robots/WAF or other live changes.

Workflow

  1. Frame one audience, channel, customer job, business outcome, market and approval boundary. Record the consequence of getting it wrong.
  2. Establish the baseline: customer questions, profile/entity consistency, content and source quality, canonical destinations, available referrals, self-report, prompt observations, platform data and (where authorised) logs.
  3. Apply the three-mode planning lens from the Carter synthesis: evergreen brand/offer facts for remembered knowledge; current sourced updates for retrieval; deep evidence, trade-offs and working for reasoning. This is a durable planning lens, not a fixed platform taxonomy.
  4. Select one content or profile slice. State its hypothesis, primary outcome, trust/cultural/accessibility guardrail, owner, time-box, stop rule and rollback path; stop the slice when its stop rule fires.
  5. Make the slice legible: who the brand is, what it does, for whom, where, under what limits, with a clear next action and an accurate canonical link. Use native channel conventions without forcing slang, hashtags or claims. Add three checks from the customer-language bank and intent map and the real-time bridge and voice contract: the outside-in customer journey, an intent/customer-language map and a recognisable human voice.
  6. Verify every current claim, source, statistic, quote, rights assertion and platform rule. Run anti-slop, creative, legal/market, language and accessibility reviews that apply to the asset. The supplied books are historical or editorial inputs, not current platform authority.
  7. Measure separately: representation, retrieval/citation observation, referral, qualified action and revenue. Record sample, date, denominator, consent and attribution limits.
  8. Check normal and failure paths, including inaccurate AI descriptions, negative or misleading UGC, broken destinations, opt-out, moderation and no-data states. Correct, quarantine or rerun the affected check; retain native adaptation, source/rights/approval handoffs, moderation and a correction path.
  9. Standardise only a demonstrated improvement in the skill, reference, template, source register, fixture or gate. Record the next re-audit.

SEO, AEO, GEO, AIO and SXO disposition (evidence-checked 2026-09-13)

Apply this disposition when a client uses the labels SEO, AEO, GEO, AIO or SXO. The labels are planning shorthand, not interchangeable platform ranking systems.

Layer What is supportable What to reject or qualify
SEO Intent-led search language, useful page content, crawlable links, indexability, technical health, page experience, and legitimate references remain the foundation. No universal quarterly refresh cadence; no promise based on “domain authority” or a fixed keyword formula. Refresh after a substantive evidence, offer, intent, or performance change.
AEO Clear answers, definitions, steps, comparisons, and natural question coverage can make content easier for people and systems to use. Never require 40–60 words. Google publishes no exact featured-snippet length and does not allow owners to mark pages for inclusion. FAQ/HowTo markup is conditional and never a guaranteed rich result.
GEO Observe retrieval, citation, claim fidelity, referral, and qualified action separately. Use original, factual, entity-clear content and authentic community distribution. “Cited in ChatGPT, Perplexity, or Gemini” is an observation, not a deliverable guarantee. Do not treat tables, headings, FAQs, E-E-A-T language, forums, or datasets as a universal recipe.
AIO Use “entity and knowledge presence” for consistent, verifiable identity and offer facts across owned and legitimate third-party sources. Define AIO before using it. Do not promise inclusion in training data, Wikipedia/Wikidata/GitHub placement, model recall, or future factual accuracy. Separate retrieval, user-triggered fetching, and potential training controls.
SXO Join the social click to a fast, accessible, intent-matched destination with a clear CTA, consent-safe handoff, and failure path; test conversion and quality. CTA/CRO improvements are not automatically SEO ranking factors. Do not report speed scores, citations, or engagement as sales without an attribution definition.

For Google, map the work back to foundational SEO and people-first content. For ChatGPT and Perplexity, verify the current search crawler and referral guidance before changing robots or WAF rules. Keep provider scope, access date, review date and uncertainty in the source register; unavailable cross-platform claims remain NOT_ASSESSED.

Outputs

Artefact Consumer Acceptance condition
AI-search visibility audit or plan Strategist, client reviewer or delivery team Audience, channel job, evidence boundary, outcome definitions, sequence, owners and gaps are explicit.
Content/profile action brief Content or community operator One real slice has channel-native copy guidance, source/rights status, CTA, moderation path and acceptance checks.
Measurement and learning record Analyst and accountable owner Prompt observations, platform data, referrals, self-report and qualified outcomes are not conflated.
Decision and gap note Approver or next workflow Unsupported, unauthorised, stale and NOT_ASSESSED items are visible with a recovery action.

Evidence Produced

Evidence Format Acceptance condition
Source and claim register Inline table or linked JSON/Markdown record Every material current claim has a verified source, scope, date, freshness, support state, uncertainty and owner.
Content/profile fact map Table Identity, offer, audience, location, proof, limits, rights and canonical destination are traceable.
Experiment record Markdown or tracker row Hypothesis, baseline, measure, guardrail, stop rule, result, rollback and standardisation decision exist.
Release review Completed gates Anti-slop, rights, legal/market, cultural, language, accessibility and approval status are explicit.

Capability and Permission Boundaries

Read and search only; analysis is read-only. Publishing, spend, live account changes, outreach and personal-data processing need explicit, action-specific client authority. Changing robots or firewall (WAF) rules, other production changes and certification claims need the same authority; edits to repository guidance are in scope for maintainers.

Degraded Mode

Without verified sources, platform access, native-language review, rights review or measurement data, return the narrowest qualified result and mark the affected checks not assessed. The legibility fact map, the SEO/AEO/GEO/AIO/SXO disposition and an experiment design can still be delivered; a missing native-language, rights or measurement check never becomes a pass.

Decision Rules

Condition Action Failure or risk avoided
The claim is current, material and supported by the source register Cite the source at the point of use and record scope, dates, freshness and limit. Stale platform or market advice.
The observation is a mention, citation, referral, sentiment or conversion Name that exact outcome; keep it separate from the others. False “AI rank” or attribution certainty.
The profile or post contains a factual, regulated, sensitive or rights-bearing claim Require owner evidence and the relevant legal/rights/market gate. Harm, rights breach or fabricated proof.
A destination is useful to people and agents Improve clear facts, accessible text, consent-safe CTA and failure path. Optimising a surface that cannot complete the job.
Evidence is partial or contradictory Narrow, quarantine or mark NOT_ASSESSED; preserve the contradiction. Confident synthesis from a weak source.
The client asks to change robots or WAF rules for ChatGPT or Perplexity Verify the current search crawler and referral guidance first and obtain owner authority. Blocking retrieval or referral by accident.
The brief includes the report's LinkedIn or citation findings Load the LinkedIn citation section of the rules reference; preserve the study scope and use the findings as test hypotheses, never as platform laws. Sample-bound findings sold as guarantees.

Quality Standards

  • British English and Uganda/East Africa defaults only where they apply; any different market, language, currency, timezone or channel reality is recorded.
  • The human job stays primary: clear answers, evidence, honest limits and a usable destination matter more than AI-facing formatting.
  • Social and community presence is part of the discoverability surface, with no buying, seeding, manufacturing or suppressing of mentions or reviews.
  • llms.txt, markdown mirrors, APIs, MCP or agent integrations appear only as a named, reversible experiment for a real consumer or task; never as a default ranking lever.
  • Each measured stage (representation, retrieval/citation, referral, qualified action, revenue) carries its own sample, date and denominator.
  • ai-marketing/anti-ai-slop ran during drafting and ai-marketing/ai-slop-audit after major iterations; an F blocks progression until fixed.

Anti-Patterns

  • Unsupported benchmark or adoption number. Fix: verify the primary source or remove it.
  • “AI rank” reported as a metric. Fix: name mention, citation, referral, action, or revenue.
  • FAQ, 50-word opening, monthly cadence, or speed target treated as universal. Fix: make it a tested local acceptance choice or remove it.
  • Inauthentic mentions, seeded comments, or manufactured reviews. Fix: use authentic, rights-cleared evidence and moderation.
  • A profile optimised without an accurate destination. Fix: trace the click, consent, form/WhatsApp path, and failure recovery.
  • A calendar presented as learning. Fix: add a hypothesis, guardrail, stop rule, and result.
  • A current platform claim copied from a book or AI answer. Fix: route it through Digital Research and mark it NOT_ASSESSED until verified.

References