The short answer
Getting cited requires three things in this order: let the right crawlers in, make your facts machine-readable and quotable, and get corroborated somewhere other than your own website. The third is the slowest and the one most programs skip, and it is the one ChatGPT weighs most heavily.
What you cannot do is run one tactic list across every engine. Only about 11% of the domains cited by ChatGPT are also cited by Perplexity, because they source information in fundamentally different ways.
They are not one audience
Treating "AI search" as a single channel is the most common and most expensive mistake in this work. ChatGPT, Perplexity and Google AI Overviews have different indexes, different freshness weighting and different trust models. A page that gets quoted constantly by Perplexity can be entirely absent from ChatGPT’s recommendations, and the fix is not more of the same.
ChatGPT
Optimizing for ChatGPT is mostly an off-page job. The structural levers that matter most are Wikipedia presence and the volume of third-party mentions of your brand, with communities such as Reddit, Quora and G2 acting as social proof it appears to weigh heavily.
What that means in practice:
- Earn mentions in publications and communities, not just links. An unlinked brand mention in a relevant thread is worth more here than a directory listing.
- Get your comparison and review presence in order on the sites in your category. This is where a model reads what other people say you are good at.
- Work toward Wikipedia and Wikidata eligibility if your organization plausibly qualifies. Do not attempt to write your own entry.
- Check your crawler policy first, which is covered below. Blocking OAI-SearchBot removes you from real-time recommendations entirely.
Perplexity
Perplexity functions as a real-time research assistant. It prioritizes fresh, well-structured data and weights content updated within the last year heavily. It is the most winnable of the three for a smaller brand, because effort and recency count for more than accumulated authority.
- Publish often, and show the date. A visible publication and last-updated date on the page is not decoration here.
- Structure in short answers. Question-shaped heading, then a complete two-sentence answer, then the detail.
- Use numbered lists and inline citations in your own content, which mirrors how Perplexity presents information.
- Exist in the communities it reads. A brand with no presence outside its own domain is hard for it to corroborate.
Google AI Overviews
AI Overviews leans on the Google index enriched with experience, expertise, authoritativeness and trust signals. This is the surface where classic SEO and GEO are least distinguishable: a page that cannot be crawled and indexed cannot be cited, full stop.
- Put a direct answer at the top of each section, before the context and the caveats.
- Use sourced figures. A number with a named source attached is dramatically more liftable than an unattributed claim.
- Keep Schema.org markup consistent across templates rather than hand-pasted per page.
- Name a real author with a visible bio and verifiable credentials, and include original data or first-hand experience where you have it.
The shared foundation
Underneath the per-engine work, four things move every surface at once.
Write claims short. Sentences that get cited in AI answers average 9.27 words with a median of 10, and the 6 to 10 word band alone accounts for 45.2% of citations. Long compound sentences are hard to lift without distorting them, so a model tends not to. Write your key facts as short standalone declaratives and put them where they can be found.
Answer first. A complete answer in the first two sentences under a question-shaped heading. Headings should mirror how people phrase a prompt, which is longer and more conversational than a keyword.
Mark it up properly. JSON-LD is what confirms entities, dates, authorship, ratings and relationships to a machine. Article, FAQPage, HowTo, Product, Organization and Person are the ones that earn their place, cross-linked by identifier so the graph resolves rather than fragmenting.
Refresh visibly. Update dates on a 30, 90 and 180 day cadence depending on how fast the subject moves, and show the date on the page.
Crawler access comes first
Before any of the above matters, check what your robots.txt permits. Every AI-visibility program we have audited that was going nowhere had at least one of these problems, and about a third had a blanket block someone added during an unrelated scraping panic.
- OAI-SearchBot is the real-time crawler behind ChatGPT search. Blocking it guarantees exclusion from real-time recommendations.
- GPTBot and ClaudeBot handle broader crawling and together account for hundreds of millions of requests a month.
- PerplexityBot and Google-Extended control the two remaining major surfaces.
These are business decisions, not defaults. Decide deliberately which you allow, write the decision down, and check quarterly that nobody has quietly changed it.
The llms.txt question
Publish one. Do not build a strategy on it. As of early 2026 no major AI company has publicly committed to reading or acting on llms.txt in production, and Google has stated it does not support it. A handful of smaller providers do fetch it, and it costs an hour to write, so it is reasonable insurance and unreasonable as a headline. Any proposal that leads with llms.txt is leading with the cheapest item on the list.
What a citation is worth
The volume is smaller than classic organic and the quality is much higher, for a simple reason: the visitor arrives pre-qualified, having already been told by a neutral-seeming source that you are a candidate.
Two figures worth knowing. Pages cited in Google AI Overviews have been measured earning around 35% more organic clicks than uncited competitors on the same results page, so the citation helps the link beside it as well. And visitors arriving from Perplexity have been measured converting at roughly eleven times the rate of traditional organic search traffic.
How to measure it
Analytics will not show you most of this, because a large share of AI-influenced visits arrive with no useful referrer: the buyer reads the answer, then searches your brand name. Measure the leading indicator directly instead.
- Build a fixed set of 100 to 300 real buyer prompts and never change it. Changing the prompt set is how programs manufacture progress.
- Run it monthly, on the same day, across every engine you care about.
- Record share of answer per engine: the percentage of prompts where you are named, and where you are not, which competitor is.
- Track branded search volume alongside it. That is where the traffic without a referrer shows up.
If a supplier cannot show you the transcripts behind that number, they are reporting a feeling.