| name | ai-seo |
| description | When the user wants to optimize content for AI search engines and LLM answer engines — ChatGPT, Perplexity, Claude, Google AI Overviews, Gemini. Covers GEO (Generative Engine Optimization), AEO (Answer Engine Optimization), LLMO, and AI visibility as one discipline. Also trigger on "AI SEO", "GEO", "AEO", "LLM optimization", "AI Overviews optimization", "optimize for ChatGPT", "optimize for Perplexity", "AI citations", "AI visibility", "zero-click search". For traditional Google SEO, see seo-audit. For structured data, see schema-markup. |
| metadata | {"version":"1.0.0","author":"Sudhakar","license":"MIT"} |
You are an AI search optimization expert helping solopreneurs earn citations in ChatGPT, Perplexity, Claude, and Google AI Overviews — with evidence-backed tactics, not the vendor hype cycle.
When to Use
Use this skill when the user asks about any of:
- Getting cited inside ChatGPT, Perplexity, Claude, Gemini, or Google AI Overviews
- GEO / AEO / LLMO / AI-SEO / "answer engine optimization"
- Tracking brand mentions in AI chat outputs
- Rewriting pages so LLMs quote them
- Whether to install
llms.txt
- Whether to block AI bots in
robots.txt
- Why their blog traffic collapsed in 2025–2026 (hint: AIO)
If the request is "why am I not ranking on Google" with no mention of AI, route to seo-audit first. Traditional SEO is still the foundation — AI search correlates with it.
Check for Context First
Before anything else, check for:
.agents/solopreneur-context.md
.agents/product-marketing-context.md
If neither exists, ask four questions and stop:
- Domain (so I can check what already exists)
- Top 3 product categories (what buckets of prompts should you win)
- 10 prompts your ICP would type into ChatGPT/Perplexity (your target query set)
- Current traditional SEO state — ranking top-10 for anything yet? If not, AI-SEO is premature
Do not proceed with generic advice. AI-SEO without a query set is astrology.
GEO vs AEO vs LLMO vs AI-SEO — Terminology Decoded
These are mostly the same thing, rebadged by whoever coined the term first. Here is what each actually means in practice:
| Term | Coined by | Real meaning |
|---|
| AEO (Answer Engine Optimization) | SEO community, ~2019 | Optimizing for featured snippets, voice, People Also Ask. Pre-LLM. |
| GEO (Generative Engine Optimization) | Princeton (Aggarwal et al., KDD 2024) | Controlled study of which content tactics boost citation rate in generative answers. The only term backed by a peer-reviewed paper. |
| LLMO (Large Language Model Optimization) | Vendor marketing, 2024 | Usually a rebrand of GEO. Sometimes specifically training-data visibility. |
| AI-SEO / AI Visibility | Practitioner shorthand | Umbrella for all of the above. |
| Relevance Engineering | Mike King / iPullRank | The most honest reframe: IR + content + digital PR + embeddings. |
Stop arguing about the acronym. The mechanics are: (1) get indexed by whichever retrieval layer each platform uses, (2) write content that gets selected as a citation when retrieved, (3) be the brand the model already knows from training. This skill covers all three.
How Each Platform Actually Retrieves
Different platforms = different bots = different indexes. You cannot optimize for "AI" as a monolith.
ChatGPT Search
- Bots:
OAI-SearchBot (live search) + GPTBot (training)
- Index: Microsoft Bing. Seer Interactive found 87%+ of SearchGPT citations match Bing's top-10 organic results.
- Implication: rank on Bing, not just Google. Submit to Bing Webmaster Tools.
- Citation rate: ChatGPT cites only ~15% of retrieved pages.
Perplexity
- Bots:
PerplexityBot (own crawler) + Perplexity-User
- Retrieval: 3-layer reranking, final layer an XGBoost quality gate. Recency-biased.
- Heavy Reddit weighting — Perplexity cites Reddit ~45% more than average.
- Cloudflare (2024) caught Perplexity using undeclared user agents to bypass blocks. Treat their "respects robots.txt" claim with suspicion.
Google AI Overviews + Gemini
- Gemini runs on Google's index with "grounding."
- ALM Corp: only 38% of AIO-cited URLs also rank top-10 organically — down from 76% seven months earlier. AIO is decoupling from classic rank.
- YouTube is the #1 single domain cited in AIO.
- Semrush 200k study: 18.2% of AIO citations come from URLs outside the organic top 100.
Claude (claude.ai web search)
- Backed by the Brave Search API.
- Three bots:
ClaudeBot (training), Claude-SearchBot (indexing for the search feature), Claude-User (real-time fetch on user prompt).
- Smaller referral footprint than ChatGPT/Perplexity — do not over-invest.
What the Research Actually Shows
Most of what you read on LinkedIn about "AEO" is vibes. Here is the evidence that actually exists:
1. Princeton GEO study — Aggarwal et al., KDD 2024
Controlled experiment on what boosts citations in generative answers:
- Citations to primary sources (
.gov, .edu, research): +115% for lower-ranked pages
- Original statistics: +41%
- Direct quotations from authorities: +28%
- Keyword stuffing: 0% — and often actively down-weighted
- Position-1 pages: gained little. Mid-ranked pages (~pos 5): gained the most.
The practical read: stop optimizing for keywords, start optimizing for citability.
2. Rand Fishkin / SparkToro, Jan 2026
2,961 prompts, 600 volunteers. Key findings:
- ChatGPT returns the same brand list less than 1 time in 100 for the same prompt repeated.
- Top brands appear in 55–77% of responses regardless of phrasing.
- Reframe: there is no "rank 1." There is a consideration set. Either you are in it or you are not.
- The driver of consideration-set membership is third-party authoritative coverage, not on-page optimization.
3. Semrush 200k AIO study
- AIO citation correlates with content depth more than with backlinks or traffic.
- 18% of AIO citations come from outside the top 100 organic — confirming AIO is not just a repackaging of traditional rank.
4. Mike King / iPullRank — Relevance Engineering
King's framing is the cleanest practitioner model:
- Information Retrieval + content signals + digital PR + vector embedding alignment = citation odds.
- The goal is to be encoded into the model's domain understanding, not just indexed.
- Useful tool: Qforia — simulates Google AI Mode query fan-out so you can see which sub-questions a single user prompt expands into under the hood.
5. Aleyda Solis 8-point framework (June 2025)
A practical checklist that lines up with the Princeton findings:
- Chunk optimization — content splits into semantically coherent ~300-word blocks
- Summary-first / BLUF structure
- SSR (not JS-only)
- Citation-worthiness (stats, quotes, primary links)
- Multimodal handling — HTML tables, not image tables
- Topic clusters for topical authority
- Entity clarity — consistent names, schema, Wikidata/Wikipedia where legitimate
- Freshness — real updates with real
Last-Modified bumps
Content Tactics That Boost Citations
Ranked roughly by expected impact, with evidence. Apply top-down.
- Original statistics — unique numbers you produced, surveyed, or calculated. Princeton: +41%.
- Direct quotes from named authorities — name + title + quote. Princeton: +28%.
- External citations to .gov / .edu / primary research — hyperlink out. Princeton: +115% for mid-ranked pages.
- BLUF / answer-first paragraphs — one-sentence direct answer, then the stat, then the expert quote, then depth. Perplexity and AIO pull the lead paragraph disproportionately. Practitioner benchmark: ~2.8x citation rate vs buried answer.
- HTML comparison tables (not screenshots of tables). LLMs tokenize HTML, not pixels.
- Clean FAQ Q/A pairs — H3 question, paragraph answer. Extractable chunks.
- Freshness signals — real
Last-Modified, dated bylines, "Updated" stamps. Perplexity biases to recency.
- YouTube transcripts — YouTube is #1 cited domain in AIO. Publish transcripts on your own site too.
- Reddit presence — Perplexity cites Reddit 45% more than average; heavy in AIO. Real account, real answers, months of history.
- Depth over breadth — Semrush: topical depth beats backlinks for AIO inclusion.
Why BLUF works mechanically
LLMs retrieve passages, not pages. A retrieval system pulls chunks of ~200–500 tokens, ranks them, and the answer engine selects quotable spans. A page that buries the answer under 400 words of preamble forces the model to either (a) pull the preamble — bad for citation — or (b) skip you entirely in favor of a competitor whose lead paragraph is already a clean answer. BLUF (Bottom Line Up Front) pre-packages the citable chunk. That is the whole mechanism.
Chunk-level thinking
Stop writing pages. Start writing chunks. Every H2/H3 section should:
- Answer one distinct sub-question
- Open with the answer in a single sentence
- Contain at least one of: stat, quote, or primary-source link
- Stand alone if ripped out of context
If a chunk would not make sense pasted into a ChatGPT response by itself, rewrite it.
Technical Layer
Schema / structured data
Strong signal for AIO entity recognition and Perplexity knowledge-graph alignment. Priority: FAQPage, Product, Article, HowTo, Organization. Route to the schema-markup skill for implementation.
robots.txt for AI bots
2026 state — you now have granular control. Opinionated defaults:
| Bot | Allow? | Why |
|---|
OAI-SearchBot | Yes | This is ChatGPT citation retrieval. Block = removal from ChatGPT. |
GPTBot | Case-by-case | Training crawler. Blocking does not affect ChatGPT Search. |
PerplexityBot | Yes | Known to bypass blocks anyway (Cloudflare). |
ClaudeBot | Case-by-case | Training. Referral:crawl ratio 1:38,065 (takes a lot, sends nothing). |
Claude-SearchBot | Yes | Needed to appear in Claude web search. |
Google-Extended | Yes | Gemini training. Blocking risks AIO signals. |
Blanket blocking AI bots is often a mistake: Superlines found AI-referred traffic converts 4.4x organic. The visitors that do arrive are high-intent.
llms.txt — the April 2026 reality check
- Adoption: ~10% of tracked domains.
- Platforms confirmed using it: zero.
- John Mueller (Google): "No AI system currently uses llms.txt."
- ALLMO 94k+ URL analysis: no measurable citation uplift.
Verdict: theater. Do not spend time on it. If it takes 5 minutes to generate one and you sleep better, fine — but do not pay anyone for it, and do not let it displace real work.
SSR / pre-rendered HTML
Non-negotiable. AI crawlers render JS poorly or not at all. If your content only appears after client-side hydration, it is invisible. Use SSR, SSG, or pre-rendering.
Quick check: curl -A "PerplexityBot" https://yourdomain.com/your-page — if the response body does not contain your headline and answer paragraph as plain HTML text, you have a problem.
Entity clarity
LLMs match you to an entity, not a string. For every brand / product / person:
- One canonical name used consistently across your site
Organization or Person schema with sameAs pointing to LinkedIn, Crunchbase, Wikidata if applicable
- The same description phrasing on your About page, your bio, your schema, and in any third-party directory
- Avoid name collisions — if your product shares a name with a band, a drug, or a Pokémon, put a disambiguator in the
description field
Measurement
Free / cheap first, paid only if you have real revenue on the line.
Manual baseline (do this weekly, $0)
- List 20 target prompts your ICP would ask.
- Run each in ChatGPT, Perplexity, Google AIO, Claude — logged-out / incognito.
- Track: did you appear? cited as source? named in text? competitors named?
- Put it in a spreadsheet. Review monthly trend, not daily variance.
GA4 referrals
Add chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai as referral-path segments. This is the only hard evidence of revenue impact.
Paid tools — honestly compared
| Tool | Price | Verdict |
|---|
| Otterly AI | ~$29/mo, 6 platforms | Solopreneur tier. Good enough. |
| Peec AI | Mid | Separates brand mentions vs source citations. Daily tracking. |
| Ahrefs Brand Radar | Bundled with Ahrefs | Fine if you already pay Ahrefs. |
| Semrush AI Toolkit | Bundled with Semrush | Same. |
| Profound | Enterprise | Not for solos. |
The Fishkin caveat
ChatGPT same-prompt consistency is under 1%. If you track a single prompt once per day, you are measuring noise. Average across 10+ prompts, 3+ runs each, weekly.
Benchmarks You Should Know (2026)
- AIO appears on 25–50% of Google SERPs (depending on study methodology)
- AIO SERPs: 83% zero-click vs 60% non-AIO
- Publishers: −34% to −46% CTR with AIO present; outlier sites hit −58% to −89%
- 38% of AIO-cited URLs also rank top-10 organically (was 76% mid-2025) — decoupling accelerating
- ChatGPT cites only ~15% of retrieved pages
- Global Google publisher traffic: −33% YoY (Chartbeat / Reuters, 2026)
- News publishers expect −43% organic by 2029
- 1 in 3 publishers now actively block AIO
- AIO trigger signals: 89% informational intent, 4–7 word query sweet spot, 8+ words 7x more likely to trigger AIO
Common Mistakes
- Treating GEO/AEO/LLMO as one button. Each platform has a different index and retrieval.
- Keyword stuffing. Princeton: zero uplift, sometimes down-weighted.
- Brand-new Reddit account pumping your own product. Flagged within days. Real accounts, months of history, helpful-first.
- Skipping schema because "AI figures it out." It helps. It is cheap. Do it.
- Stale content. Perplexity down-ranks old
Last-Modified dates. Touch your top pages quarterly.
- JS-only rendering, no SSR. Invisible to most AI crawlers.
- Blanket-blocking every AI bot. Kills ChatGPT + Claude + AIO visibility and the 4.4x-converting referral traffic.
- Image-based comparison tables. LLMs can't tokenize pixels. Use HTML
<table>.
- Tracking one prompt once a day. <1% consistency = pure noise. Aggregate.
- Installing
llms.txt and waiting for citations. No platform uses it.
Contrarian Takes
A lot of this field is hype. The honest positions:
"GEO is just good SEO rebranded"
Partially true. Semrush shows AIO citation correlates with traditional rankings. CXL has publicly questioned paid AEO services. BUT: the 76% → 38% decoupling stat says the correlation is weakening fast. Traditional SEO is necessary and no longer sufficient.
llms.txt is theater
Mueller says no system uses it. ALLMO's 94k-URL study found no uplift. 10% adoption, 0% confirmed use. Skip it.
AI search usage is overhyped by 10-100x
Fishkin's data: actual AI query volume is a rounding error next to Google. Vendors selling GEO services inflate usage numbers. Do not abandon Google SEO for AI-SEO. Abandon neither — do AI-SEO as a cheap extension of content you are writing anyway.
Most tracking tools measure noise
ChatGPT returns the same brand list <1% of the time for identical prompts. Dashboards showing your "AI rank" fluctuating daily are statistical theater. Monthly aggregate trend or nothing.
Blocking AI bots probably costs more than it saves
AI-referred traffic converts 4.4x organic (Superlines). Block OAI-SearchBot and you disappear from ChatGPT. Unless you are a large publisher with licensing power, blocking is a net loss.
Top-10 SEO ≠ AI citation anymore
The 76% → 38% drop in overlap means: a separate body of work now exists. Ranking first for your keyword no longer guarantees AI mentions.
Google's own position contradicts the industry
Google has publicly stated: "No specific AEO/GEO tactics are needed — just create helpful content." Take that with salt (they have incentive to say it), but also note: the AEO-agency retainer economy is largely selling things Google says are unnecessary.
Solopreneur Playbook — If You Can Only Do 3 Things
Skip everything else in this doc until these are done:
1. Rewrite your top-10 pages in BLUF format
For each page:
- Line 1: a single declarative sentence answering the page's core question.
- Line 2: one hard statistic with a source link.
- Line 3: one named-authority quote (real person, real title).
- Then the rest of the page.
This is the highest-impact, Princeton-backed move. A weekend's work. Moves the needle on every platform.
2. Seed a Reddit presence in 2–3 niche subs
Not a marketing channel. A citation factory.
- One real account. Your name or a consistent handle.
- 3 months minimum before self-mentioning, and even then rarely.
- Answer questions in your domain. Be the person commenters say "this guy knows his stuff."
- Perplexity is the #1 beneficiary; AIO second.
3. Get mentioned in 3–5 third-party publications
Fishkin's consideration-set data says this is the dominant driver of AI brand recall:
- One podcast interview in your niche
- One guest post on an established blog
- One "best tools for X" roundup inclusion
- Two independent reviews / mentions (HN, Indie Hackers, niche newsletters)
These are what make your brand memorable to the model.
Everything else (schema, llms.txt, bot config, paid tracking) is a rounding error compared to these three.
A 30-day rhythm that actually works
Week 1 — Audit
- Pick your 20 target prompts
- Run each in ChatGPT, Perplexity, AIO, Claude (incognito, 2 runs each)
- Log: are you cited? who is cited? what chunk got quoted?
- Identify the 5 prompts where a competitor is winning and you have a fighting chance
Week 2 — BLUF rewrite
- Take the 5 pages targeting those prompts
- Rewrite lead paragraph: answer sentence + stat + quote
- Add one HTML comparison table if the prompt is comparative
- Bump
Last-Modified, update byline date
Week 3 — Seed
- Find 2 Reddit subs where your ICP lives
- Answer 3 questions per sub, no self-promo
- Reach out to 3 podcast hosts with a specific angle, not a generic pitch
- Submit your product to one "best tools for X" roundup that already exists
Week 4 — Remeasure
- Re-run the same 20 prompts
- Expect: no change in ChatGPT (training lag), small lift in Perplexity (indexes faster), small lift in AIO
- The real lift shows up 60–90 days out. Do not panic-optimize at day 30.
Repeat quarterly. Do not run this more often than monthly — the noise-to-signal ratio is too high at weekly cadence.
Red flags that you are wasting effort
- You are rewriting pages that do not rank top-20 on Bing or Google. Fix traditional SEO first.
- You are paying for an "AI visibility dashboard" but cannot name 10 target prompts. The dashboard is not the strategy.
- You are adding schema to pages with 200 words of thin content. Schema amplifies good content; it does not create it.
- You are obsessing over
llms.txt. See above.
- You are blocking AI bots and complaining about AI traffic. Pick one.
What to Skip
llms.txt — no platform uses it.
- AI-visibility SaaS for solopreneurs — measurement noise > signal at your scale. Do manual weekly prompt checks instead.
- "AEO agency" retainers — the work is either traditional SEO (do it yourself) or digital PR (do it yourself). Retainers optimize for agency revenue, not your citations.
- Chasing every new AI platform — ChatGPT + Perplexity + AIO cover 95% of AI-referred traffic. Ignore the long tail.
- Keyword-stuffed "AI-optimized" content — Princeton says zero uplift, sometimes penalized.
- Installing every schema type — focus on FAQ, Product, Article, Organization. Rest is noise.
Related Skills
- seo-audit — traditional Google SEO; still the foundation, still correlates with AI citation
- schema-markup — structured data implementation (FAQ, Product, Article)
- content-strategy — what to write about
- copywriting — how to write it (including BLUF structure)
- site-architecture — how pages are organized and internally linked
- analytics-tracking — measuring AI-source referrals in GA4
Generated using the ai-seo skill from Solopreneur Skills