Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Get recommended by AI search (ChatGPT, Perplexity, Google AI Overviews, Gemini, Claude) using GEO / AEO: audit per-model visibility, structure content so LLMs cite it, earn off-site mentions and entity consistency, then run it as a monthly cadence with GA4 attribution. Use whenever the user wants to appear in AI answers, asks about GEO, AEO, "generative engine optimization", LLM visibility, ChatGPT/Perplexity recommendations, AI Overviews, being cited by AI, "why does ChatGPT recommend my compet
.claude/skills/whatsuppiyush-ai-search-geo/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 147% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 210% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 259% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 269% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 169% | 0% |
The discipline of getting named and cited by AI answer engines. It is mostly a brand-positioning problem, not a technical one: roughly 85% of what AI says about you comes from off-site. You win by measuring per-model visibility, formatting content to be extractable, engineering consistent off-site signals, and running it as a continuous loop with attribution that can actually see AI-referred traffic.
for high-intent "best for X", comparison, and alternatives prompts.
use seo-content. Classic ranking still matters here (a #1 Google result is ChatGPT-cited ~43% of the time), so do not skip it.
copywriting-messaging. This skill tells you the structure and placement; the words are a copy job.
analytics-data. This skill gives the custom-channel regex and what to expect; the measurement plumbing is analytics.
Follow the four-phase method: Diagnose -> Position -> Build -> Scale. Then maintain it.
sharply (only ~11% domain overlap between ChatGPT and Perplexity), so test every major model. Run the query x model audit, log brand appearance, direct links, and ghost citations, then sort every gap into one of three types: coverage (rivals have multi-source reinforcement you lack), consistency (your messaging contradicts itself across sources), or specificity (you are described broadly while a rival owns a niche).
authority) and manage presence across the surfaces that shape answers (reviews, communities, publications, profiles), weighting toward the off-site 85%.
4+ channel distribution) and set up attribution, because GEO shows a 4-8 week lag and mostly hides in "direct".
Prioritize high-intent queries (comparisons, "best for X", alternatives) first everywhere.
entity, prompt mapping, and monitoring workstreams
60-minute top-10-pages diagnostic, then classify gaps as coverage / consistency / specificity
first, comparison tables, statistics, human-written
consideration set (85% of mentions are third-party)
Wikidata/Reddit, plus ecosystem presence on YouTube/Reddit
-> topical authority), multi-channel distribution, and GA4 attribution for hidden AI traffic
listicles/comparisons plus Reddit/Quora. Comparative listicles are 32.5% of AI citations (listicles overall ~40%); for pro services, 80.9% of cited list content came from neutral third parties.
5-10 unbranded buyer questions across ChatGPT / Perplexity / Gemini / Claude; re-run 15-20 questions monthly (~2-3 hrs).
adding statistics ~41%; lead each section with a 40-60-word self-contained answer; 82% of AI-cited articles are human-written.
brand web mentions 0.664. Content across 4+ channels is 2.8x more likely to appear in ChatGPT. Baseline with a free AEO grader (0-100).
chatgpt.com|perplexity.ai|gemini.google.com|copilot.com|claude.ai;expect a 4-8 week publish-to-citation lag; ~70% of AI visitors arrive with no referrer (look like direct). AI-referred visitors convert ~11x organic; AI mentions convert 7.1% (vs 7.8% paid search).
time; inline brand-homepage links in ChatGPT answers jumped 0.4% -> 6.2% (May 7, 2026).
3.2x AI-referred traffic in 60 days via a documentation rewrite. AI search is projected to eclipse traditional search by 2028.
Every play above, in full.
AI search optimization (GEO) is mostly a brand-positioning problem, not a technical one: the headline stat is that roughly 85% of what AI says about you comes from outside your website. You influence AI answers by engineering signals across the third-party sources models trust, tracked continuously. The method is four phases: Diagnose → Position → Build → Scale.
When buyers in your category are asking AI tools (ChatGPT, Perplexity, Google AI Overviews) for recommendations and you want to be named, especially for high-intent, pipeline-driving prompts.
Becomes a "GEO / AI-search visibility" SKILL.md: runs prompt-set diagnostics across LLMs, maps buying prompts, and outputs a Position/Build/Scale action list weighted toward off-site authority sources.
Before you can improve AI-answer visibility you have to measure it, per-model, because the models disagree sharply about who to recommend. The audit is a query×model grid of unbranded buyer questions; the diagnosis sorts every gap into one of three fixable types. This is the concrete "Diagnose" step under ai-search-optimization.
content-structure-for-ai-citations, off-site-and-listicle-citations, entity-consistency-and-ecosystem.Becomes an ai-visibility-auditor: input = brand + top pages + competitor set; output = a filled query×model grid with brand-appearance/ghost-citation flags, a gap classified per query (coverage/consistency/specificity), and a ranked fix list weighted to high-intent queries.
LLMs extract and cite passages, not pages. So the on-page work in GEO is formatting content into self-contained, extractable chunks the models can lift verbatim. Each structural choice below carries a hard citation-lift number; this card is the on-page "clarity" lever (the off-site lever lives in off-site-and-listicle-citations).
entity-consistency-and-ecosystem).Becomes a citability-formatter: input = a draft/page; output = the page re-chunked into 120-180-word sections, each opening with a 40-60-word self-contained answer, comparisons converted to tables, and a flag on missing statistics.
Models recommend brands they're confident about, and confidence comes from a single, consistent story repeated across the surfaces they trained on. So GEO has an entity layer: score your baseline, make your positioning identical everywhere (site, LinkedIn, G2, Crunchbase, Wikidata), and seed authentic mentions on the platforms that correlate most with AI recommendations. This is the "consistency" fix from the audit, plus the ecosystem-presence fix.
off-site-and-listicle-citations (earn the mentions), this card makes sure those mentions all tell the same story.Becomes an entity-consistency-auditor: input = your profiles (site/LinkedIn/G2/Crunchbase/Wikidata/Reddit) + positioning; output = an AEO baseline score, a diff of inconsistent descriptions to fix, and a ranked ecosystem-seeding plan weighted to YouTube/Reddit/newsletters.
GEO isn't a project, it's a recurring loop: monitor how the models describe you, publish answer pages on a steady beat, distribute across the surfaces models learn from, and measure the AI-referred traffic that mostly hides as "direct" in GA4. This card is the ongoing operating rhythm plus the attribution setup to prove it.
chatgpt.com|perplexity.ai|gemini.google.com|copilot.com|claude.ai. Expect a 4-8 week lag between publishing and citation.Becomes a geo-cadence-runner: input = buyer question set + content backlog; output = a monthly recon checklist (queries × models × tool), a biweekly publishing calendar in the right format order, a GA4 custom-channel regex, and a distribution plan hitting 4+ channels.
The single biggest GEO lever is off your own site: models mostly repeat what third parties say about you. So the work is getting named in the listicles, comparison roundups, and community answers that AI models pull from, and getting into the "consideration set," because being in it more than doubles your citation rate.
entity-consistency-and-ecosystem (be consistent across those third-party surfaces) and content-structure-for-ai-citations (make what you publish extractable).Becomes an off-site-citation-planner: input = category + target queries; output = a target list of listicles/roundups/communities to get into, a vs-page build plan, and a consideration-set checklist ranked by citation-share potential.
From God of Skills: a curated, hand-tested directory of AI skills, prompts, templates and image style guides. Source: https://godofskills.com/skills/ai-search-geo?ref=claude-skill
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 28,346 | 22,939 | -19% | 1 | 1 | 0% | 5,472 | 9,843 | +80% | 0 | 0 | — |
case-02 | fail→pass | 21,231 | 16,338 | -23% | 1 | 1 | 0% | 3,441 | 8,493 | +147% | 0 | 0 | — |
case-03 | fail→fail | 27,177 | 19,451 | -28% | 1 | 1 | 0% | 5,821 | 9,576 | +65% | 0 | 0 | — |
case-04 | pass→pass | 7,651 | 12,742 | +67% | 1 | 1 | 0% | 1,453 | 7,986 | +450% | 0 | 0 | — |
case-05 | pass→pass | 15,826 | 14,101 | -11% | 1 | 1 | 0% | 3,139 | 8,073 | +157% | 0 | 0 | — |
case-10 | fail→pass | 12,818 | 7,963 | -38% | 1 | 1 | 0% | 2,236 | 6,939 | +210% | 0 | 0 | — |
case-06 | pass→pass | 17,898 | 19,437 | +9% | 1 | 1 | 0% | 3,973 | 9,705 | +144% | 0 | 0 | — |
case-07 | fail→pass | 13,383 | 8,878 | -34% | 1 | 1 | 0% | 2,074 | 7,452 | +259% | 0 | 0 | — |
case-08 | fail→pass | 9,740 | 4,024 | -59% | 1 | 1 | 0% | 1,702 | 6,282 | +269% | 0 | 0 | — |
case-09 | fail→pass | 11,815 | 3,991 | -66% | 1 | 1 | 0% | 2,347 | 6,303 | +169% | 0 | 0 | — |
case-11 | fail→pass | 11,738 | 6,509 | -45% | 1 | 1 | 0% | 1,915 | 6,611 | +245% | 0 | 0 | — |
case-12 | fail→pass | 10,730 | 10,795 | +1% | 1 | 1 | 0% | 1,919 | 7,590 | +296% | 0 | 0 | — |
case-13 | fail→pass | 11,735 | 5,633 | -52% | 1 | 1 | 0% | 2,274 | 6,576 | +189% | 0 | 0 | — |
case-14 | fail→pass | 11,345 | 3,591 | -68% | 1 | 1 | 0% | 1,969 | 6,041 | +207% | 0 | 0 | — |
case-15 | fail→pass | 15,888 | 5,799 | -64% | 1 | 1 | 0% | 2,920 | 6,626 | +127% | 0 | 0 | — |
case-16 | fail→pass | 13,082 | 9,595 | -27% | 1 | 1 | 0% | 2,426 | 7,498 | +209% | 0 | 0 | — |
case-17 | fail→pass | 13,334 | 3,129 | -77% | 1 | 1 | 0% | 2,387 | 6,039 | +153% | 0 | 0 | — |
case-18 | fail→pass | 11,848 | 5,275 | -55% | 1 | 1 | 0% | 2,150 | 6,493 | +202% | 0 | 0 | — |
case-19 | fail→pass | 11,579 | 11,002 | -5% | 1 | 1 | 0% | 2,081 | 7,533 | +262% | 0 | 0 | — |
case-20 | fail→fail | 11,809 | 12,007 | +2% | 1 | 1 | 0% | 2,111 | 7,597 | +260% | 0 | 0 | — |
case-21 | fail→pass | 9,566 | 5,257 | -45% | 1 | 1 | 0% | 1,987 | 6,553 | +230% | 0 | 0 | — |
case-22 | fail→pass | 10,818 | 1,871 | -83% | 1 | 1 | 0% | 1,983 | 5,856 | +195% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +73 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.