AI search visibility: how to get cited by ChatGPT, Claude, and Perplexity in 2026

Between 20% and 40% of high-value queries no longer happen on Google but on ChatGPT, Perplexity, Gemini and Google's own AI Overviews. The user asks a question, the model answers in prose and cites 3 to 8 sources. If your SMB is not one of those sources, you stay invisible even if you rank #1 on Google.

Models combine three signals: domain authority (how often the site is cited in the training corpus and the open web), answer clarity (explicit questions with short direct answers on the page) and freshness (lastmod, visible date, real changes). Pages with FAQPage schema, an llms.txt and structured prose carry more weight than long-paragraph blogs.

No. Traditional SEO is the foundation of AEO (Answer Engine Optimization). If your page does not rank on Google, the models almost never find it. Keep everything from SEO: technical, useful content, backlinks, schema. On top of that, add direct-answer structure for AI.

Wikipedia, Reddit, Stack Exchange, institutional sites (gov, edu) and a rotating set of high-authority media outlets lead the citations. For niches, models also pull specialist blogs with proprietary data, long-form guides with clear structure, and verifiable reviews. Earning a mention with a backlink from one of those citable sources sharply raises your odds of being cited directly.

Use every H2 as an explicit question ("How does Google Ads work for electricians in Medellín?"), put a 1 to 3 sentence answer right below in clean declarative prose (the part the model lifts verbatim), and develop the long-form content under that extract. Mark each Q+A with Question/Answer microdata and mirror the list in a FAQPage JSON-LD block that matches the on-page wording exactly.

At minimum: Organization or Organization + ProfessionalService on the home, Article + BreadcrumbList + FAQPage on every guide, LocalBusiness if you serve a physical location and Product or Offer if you sell online. Use inline JSON-LD in the head, not microdata alone, and make sure the FAQPage schema text matches the on-page text exactly (Google penalizes mismatches).

E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) weighs more every year in how Google and AI models judge a source. For an SMB, that means signing articles with a real author, showing experience (years, credentials, cases), linking to primary sources, and exposing verifiable contact and legal information.

AthenaHQ, Profound and Otterly track when and how brands appear in ChatGPT, Perplexity and Gemini answers. Data is still limited and expensive. For an SMB, a weekly manual check is enough: ask 5 to 10 real questions about your niche in ChatGPT and Perplexity, note whether you are cited and which sentence does the citing.

The most common mistakes: stuffing pages with keywords instead of real answers, accidentally blocking AI crawlers in robots.txt, hiding answers behind JavaScript-rendered accordions, not signing articles, not publishing llms.txt, skipping FAQPage schema and leaving pages with no visible date.

Month 1: publish llms.txt at the root, add Organization/ProfessionalService schema on the home and Article + BreadcrumbList on every guide. Month 2: restructure your 3 to 5 highest-traffic pages with each H2 as a question + 1 to 3 sentence answer + Question/Answer microdata + matching FAQPage JSON-LD. Month 3: run weekly manual visibility checks, sign every article with a real author, and earn 2 to 4 fresh mentions from citable sources.