#1
From unmentioned to the top recommendation
Playbook
AI visibility
AI visibility is MAPS applied to the machine reader. Systems start with learned associations, fan out into related searches, choose a subset of sources, extract passages, compare claims and decide what to mention or cite. Citations are the output; the four conditions are the inputs. This playbook keeps the specificity the SEO and GEO community needs, inside the same framework the rest of marketing runs on.
When to use it
The brand is described wrongly, cited rarely or recommended never in ChatGPT, Google AI Overviews and AI Mode, Perplexity, Claude or Copilot.
Where to start
Usually Meaning or Presence. Save ten brand-description answers across platforms and compare them with the canonical identity; if they disagree, start with Meaning. If they agree but competitors are cited where you are not, start with Presence.
The work
Define. Prove. Distribute. Enable.
Define
- Canonical identity sentence and fact register; fix contradictions across owned pages and external profiles first.
- Prompt universe by audience and journey stage: category, problem, comparison, recommendation, alternatives, use cases, trust, transaction.
- Entity and relationship map with disambiguation for colliding names.
- Set the outcome before the platform: accurate description, better answers to priority questions, qualified visits, or subscriptions.
Prove
- Original research, documented outcomes and tested experience that answer a real need, with methods, dates and sources beside the claims.
- Real authors with relevant expertise; no invented credentials.
- Fair, criteria-stated comparisons; disclose conflicts; no self-awarded best.
- Trace repeated mentions to their source: syndicated or commonly owned pages are not independent corroboration.
Distribute
- Core and fan-out query coverage for priority prompts.
- Presence in the third-party sources systems repeatedly cite: publishers, reviews, communities, directories, marketplaces, partners.
- Competitor source gaps, closed only where the source is relevant to the audience.
- Repeat by market, language and platform where the ecosystem changes.
Enable
- Verify crawl, index, render and snippet eligibility; check search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot) separately from training bots and user-triggered fetchers.
- Answer-first, self-contained sections with named entities and scope; descriptive headings and accessible tables; no fixed word-count rule.
- Schema and identifiers that reinforce facts already correct on the page; defer llms.txt or Markdown unless a documented consumer needs it.
- Truthful publication and revision dates; refresh substance, not timestamps.
Worked example
From unmentioned to recommended.
One brand, one category, two weeks. Most of the work was words and facts, plus digital PR.
11%
Share of AI answers in the category
2 wks
Of work, most of it words and facts, plus digital PR
Meaning
One sentence, one category.
Authority
Proof on the page.
Presence
The sources that mattered.
Signals
Facts machines could read.
Measure
Use a baseline, not a vibe.
A documented prompt panel; repeat comparable observations and retain prompt, model, mode, time and context. Separate eligibility, retrieval, citation, mention, recommendation and business outcome; define every denominator including null results. Use Google's generative-AI reporting and Bing's AI Performance within their documented limits and never merge them with sampled answers or analytics sessions. Test with baselines and unchanged comparison pages.
Caution
Structured data, llms.txt, fixed answer blocks, timestamp-only freshness and review volume are not demonstrated citation levers. Treat lab results and vendor case studies as hypotheses for bounded tests.
Run the audit first.
The playbook assumes you know the constraint. The 24-statement audit takes ten minutes and tells you which pillar and which reader to start with.