One framing point up front, because it shapes the whole article: the seven steps do not all work at the same speed. Perplexity searches the web live for almost every request and therefore reacts quickly to structural changes on your website. ChatGPT leans more on trained knowledge: mentions build up here over months, not weeks. Anyone who wants to keep both systems in view needs patience with the slow levers and should not be misled by fast progress on the others.
Why this is more than an SEO buzzword
When people hear “AI visibility”, many first think of SEO under a new name. That falls short. A search engine weighs relevance signals, search terms, backlinks, technical hygiene, and returns a list a user picks from. An AI system, by contrast, writes an answer and names a handful of brands in it, or none. What sits behind the term and how it differs from classic SEO is explained in detail in What are GEO and AEO?.
For practical purposes this framing is enough: a shop can sit on page one of Google and simply be missing from the ChatGPT answer for the same category, because its content is optimised for search engines but hard for a language model to use. The other way round, a brand with a moderate ranking can be named because many independent sources describe it clearly. The following seven steps aim at that second case: making sure a language model finds something usable about your brand at all.
They fall into two groups. The first three are technical: they make sure the automated readers of AI systems can read and classify your site. The remaining four are editorial: they make sure that what is read is actually worth quoting.
The technical foundation: the entry ticket for AI bots
Without the following three points, even the best positioning has no technical entry ticket. They take little effort and are entirely in your own hands.
1. Schema.org: mark up Organization and Product completely
Schema.org is a vocabulary a website uses to mark up facts about itself in machine-readable form: as JSON-LD, invisible in the source code but directly usable by search engines and AI systems. Two entities are central for a brand: Organization (name, logo as a complete image object, contact details, sameAs references to official profiles such as LinkedIn) and Product (name, description, brand, price, availability and, where genuinely present, a review summary).
In practice, standard themes often leave a gap: sameAs is missing, the logo is only a URL instead of an ImageObject, AggregateRating is missing even though real customer reviews are visible on the page. You can check this in a few minutes: open the page source and search for application/ld+json, or use a free testing tool such as Google’s rich results test.
2. llms.txt: a signpost, honestly assessed
Next to robots.txt there is a much younger proposal: llms.txt, a file in the root directory that lists in plain Markdown which pages are particularly relevant for an AI system. Part of the honest picture: the specification is not a formally adopted standard, studies of actual usage so far show low direct access numbers, and no major AI provider has publicly confirmed it as a ranking factor. That is no reason to leave it out, the file costs nothing and some agent tools already use it, but it is not a main channel. Anyone setting it up should treat it as a foundation, not as a growth lever.
3. robots.txt: allow the AI bots explicitly
Most websites have a working robots.txt for classic search engines. What is often missing: explicit permission for the AI-specific bots that now travel with their own user agent: GPTBot (OpenAI), PerplexityBot (Perplexity), ClaudeBot (Anthropic), Google-Extended (Google’s AI training, separate from the regular Googlebot), and others. If a bot is not explicitly allowed, some systems read that as an exclusion to be safe. The fix is a block with explicit Allow lines at the end of the file, added in a few minutes. One point of context: this opens nothing that is not publicly visible anyway. It only makes unmistakably clear that reading is permitted.
The editorial foundation: what an AI system can actually quote
The technical entry ticket alone does not fill an answer. The following four steps make sure that what an AI system finds on your site and about your brand is suitable as an answer in the first place.
4. A quotable definition sentence in the visible text
Studies of how language models extract statements show that short claims with concrete content, clearly assigned to one brand, are paraphrased more readily. Most website home pages instead open with a promotional header line: strong in tone, weak in concrete information. Add a neutral sentence directly under the headline or as the first paragraph, one that would work as an AI answer, for example in the form: “[Brand] is a [category] provider from [region] that [concrete promise in half a sentence].” Leave out marketing adjectives: a language model treats promotional sentences as noise and extracts them less well than factual statements.
5. FAQ in the wording of real buying questions
Most FAQ sections answer questions a company asks itself: “how does delivery work?”, “which payment methods are available?”. Useful, but rarely what a customer actually types into ChatGPT or Perplexity. Deliberately add questions in the wording customers use, “which shop for [category] is worth recommending in Germany?”, “what is a good alternative to [well-known competitor]?”, and answer them briefly, concretely and without sales language. That serves the answer logic behind answer engine optimisation directly: content that answers a concrete question cleanly is picked up more readily by language models.
6. Catalogues and directories as third-party sources
The most underestimated lever: where a brand is mentioned often counts for more than what stands on its own website. Industry directories, comparison portals, trust-seal profiles and software catalogues, depending on the industry for example OMR Reviews, Capterra, Trusted Shops or a relevant trade directory, are the kind of third-party source language models assemble their picture of a brand from. Studies of how AI systems cite sources suggest that the frequency of a brand mention across several reputable, independent sources is a stronger predictor of an AI citation than the sheer number of backlinks. One complete directory entry therefore often does more than another blog article on your own domain.
7. Press and community mentions with attribution
The slowest but, in the long run, most important lever, especially for the trained knowledge behind ChatGPT. Two forms work particularly well: named quotes with attribution, a sentence with name, role and company instead of an anonymous “we think that …”, and genuinely helpful contributions in trade forums or communities where a brand is mentioned naturally because it really fits the question. Both take more time than the technical points above, but they are exactly the signals that turn a brand from a website presence into a brand actually described across the web.