AI search optimization (AEO + GEO) for crypto and Web3 brands

The output differs because the surface differs. Google SEO targets the 10 blue links on a SERP; AI search optimization targets the source paragraphs that AI tools quote inside an answer. When ChatGPT search answers "best crypto licensing law firm" with a sourced list, the goal is to be one of the sources cited.

The work splits across three tracks. AEO — Answer Engine Optimization — restructures pages so AI extraction layers can pull self-contained answer blocks: H1 disambiguator, Quick Facts table near the top, H2 questions with first-sentence direct answers under 30 words. GEO — Generative Engine Optimization — produces new content built to be cited (specific numbers, dated claims, named experts, primary-source links). Infrastructure — schema.org Person with sameAs verification, FAQPage with speakable, llms.txt, AI-crawler robots.txt — tells AI systems how to read and trust the source.

One concrete pattern from 2024–2025: a Series-A RWA tokenization client (NDA, 8 months) earned 44 confirmed AI citations across Perplexity, Phind and ChatGPT after we ran the AEO restructure on 6 priority pages plus the documentation site. Inbound demo requests went +108%. Google rankings barely moved during the same period because the buyer was researching inside AI tools, not in classic search.

The prompt list is built during discovery — 22 commercial and consideration-stage queries selected from your ICP, current sales conversations and our crypto-specific prompt corpus. Each prompt runs weekly across ChatGPT (gpt-4o, gpt-4.5), Perplexity, Gemini, Claude and Google AI Overviews. We capture the full answer, the sources cited, the position of each source in the answer, and the brand mentions in unbranded category queries.

The 22-count was tuned across 14 engagements. Below 15 prompts, the monitor misses category drift. Above 35 prompts, weekly hand-review becomes shallow and the analyst starts marking citations without reading the surrounding answer context, which is where the regulatory-copy risk actually lives. 22 is the operating sweet spot.

Output of the monitor is the monthly citation report — counts per platform, prompt-by-prompt share-of-voice against three named competitors, citation-quality scoring (was the cite favorable, neutral, or comparative-against-a-competitor), and the trend deltas week-over-week.

Three properties carry most of the weight in YMYL verticals. sameAs — at least two independent profiles per named expert (LinkedIn plus a conference talk, regulator submission, or authored publication). knowsAbout — specific topical fields the expert is associated with. jobTitle and worksFor — establishing the institutional anchor.

We measured the byline effect through a controlled experiment in early 2025 on one crypto-licensing client's blog. 12 pieces under "Editorial Team" byline, 12 equivalent pieces under a named expert with full schema.org Person markup. The named-byline pieces averaged 1.6× the AI-citation rate over a 60-day post-publication tracking window. Through one algorithmic event during the period, the named pieces retained ~80% of citation rate while the anonymous pieces dropped to ~50%.

FAQPage schema with speakable extension on cost, timeline and qualification questions earns rich-result eligibility within 30–60 days when the rest of the markup graph is clean. ItemList with Person references is the correct pattern for ranking pages (CryptoLawIndex-style) where naive Organization-per-firm markup creates entity confusion.

Any agency promising "guaranteed 10 AI citations per month" is selling a number disconnected from how AI systems pick sources. Three things are stochastic: the model version (vendors silently push updates), the index snapshot (Perplexity reindexes weekly, Gemini differs), and the prompt phrasing (different phrasings of the same intent hit different source sets).

What we commit to is process — the weekly monitor, the structural rollout cadence, the named-expert byline coverage, the schema audit before every release, the post-launch fetch re-verification on donor placements. When the process is held tight, citation rate compounds: across our 14 engagements 2024–2025 the median growth was 1.4–1.8× per quarter after month three.

Two clients we worked with through full algorithmic events. The one that retained tight editorial workflow held citation rate within 90% of pre-event. The one that dropped editorial discipline mid-quarter (founder-driven pivot, content review skipped for six weeks) lost 60% of citation rate inside two months and took four months to recover. Process discipline is the lever.

AI tools pull source paragraphs verbatim or near-verbatim. When ChatGPT quotes a sentence from your site that says "guaranteed APY" or "regulator-endorsed", that sentence is now appearing in an AI answer with attribution back to you. MiCA Article 66 and the UK FCA financial-promotion regime both treat the original publication as the responsible party.

Our regulatory review workflow runs before publication for every YMYL piece. Compliance lead — your in-house compliance lead or our contracted reviewer — passes 2–4 days per piece flagging the four banned-claim categories (guaranteed returns, risk-free framing, implied regulator endorsement without specific licence reference, universal-suitability suggestions). For MiCA-bound clients we maintain a per-jurisdiction copy-rules document so the German variant satisfies BaFin and the French variant satisfies AMF.

Practical consequence: AI-search visibility and MiCA compliance are not in tension — they reinforce each other. Specific numbers, dated claims, primary-source links, named-expert bylines are all things AI search systems weight as authority and that MiCA Article 66 effectively requires. Vague AI-default copy fails both filters.