AI Search Visibility: How to Show Up in ChatGPT and Perplexity

A prospect asks ChatGPT or Perplexity for a recommendation in your category, gets three names, and yours is not one of them. You check your own site, it looks fine, and nobody can tell you why. We had the same experience with our own company, so we went and found out how these systems choose their sources. The short version: they do not remember companies, they search for them, and they can only recommend what they can find and corroborate. This article is what we learned and what we changed, written for B2B service companies rather than for publishers.

How AI answer engines pick their sources

When you ask a current assistant a question with a commercial or factual answer, it does not answer from training memory alone. It runs a web search, reads a handful of results, and composes an answer with citations. Perplexity does this on every query; ChatGPT does it whenever the question calls for current information. The practical consequence is that AI visibility is downstream of search visibility. If the search step returns nothing about you, the model has nothing to cite, and it will not invent you.

The second consequence is subtler. These systems reason about entities, not pages. To decide that “Fidelisa Veda” is a software agency in Serbia serving US clients, the model wants to see that same claim on several independent sources: your site, a business profile, a professional network, a directory, a review platform, an article somewhere. One source cannot corroborate itself. A company that exists only on its own website is, to an answer engine, an unverified claim.

Structure helps the reading step. Pages that state plainly who the company is, what it does, where it operates and what a service includes, in text and in structured data, are easier to extract from than pages built of slogans. Question-and-answer blocks are particularly useful, because they match the shape of what users ask.

What we did on our own site

We describe this as practice, not as a guarantee, because nobody can promise a placement in an AI answer any more than a position in Google. First, we made the entity consistent: the same company name, description, address country, phone and email on the site, in the Organization structured data, and on every social and business profile. The structured data carries sameAs links to those profiles so the identity chain is machine-readable.

Second, we added structured data to every page type: Organization and ProfessionalService for the company, Service with a price specification for each offering, BlogPosting for articles and FAQPage wherever there is a question-and-answer block. Third, we published an llms.txt file, a plain-text summary of who we are and what we offer, written for machine readers, and our robots file names the AI crawlers explicitly rather than blocking them by accident.

Fourth, we wrote FAQ blocks on every service page and article, with questions phrased the way buyers ask them and answers that open with the direct answer. Fifth, we connected the site to search console tools, submitted the sitemap and started earning independent mentions. That last step is the slow one and the one that matters most, and it is not a code change at all. A website rebuild can set all of this up in a week; the corroboration takes months.

Why one source is not enough

Here is the trap for a small B2B firm. You can have a technically perfect site, with every schema type correct and every page indexed, and still be invisible to answer engines, because the model has no second source to check your claims against. This is not a penalty. It is the same skepticism a careful buyer applies: a company that only says good things about itself has not yet been vouched for.

Corroboration comes from places you do not fully control: a client who mentions you in a case study, a directory listing with the same description, a podcast appearance, a guest article, a partner page, a review with your company name spelled the same way. Each one is a small vote that the entity is real and does what it says. The work of AI visibility, for a service company, is mostly the work of collecting those votes.

A practical 30-day checklist

Week one, consistency. Write a two-sentence company description and use it everywhere: site footer, About page, every social profile, every directory. Check that name, country, phone and email match exactly. Add or fix Organization structured data with sameAs links to each profile. Verify the site in Google Search Console and Bing Webmaster Tools and submit the sitemap; Bing matters because ChatGPT's search runs on it.

Week two, structure. Add FAQ blocks with FAQPage schema to each service page, using real questions from sales calls. Make sure each service page states in plain text what the service includes, who it is for and what a typical engagement looks like. Publish an llms.txt. Confirm robots.txt does not block AI crawlers. Week three, corroboration. Claim and complete your business profile and the two or three directories relevant to your category. Ask two clients for a written testimonial you can publish and they can publish. Pitch one guest article or podcast in your niche.

Week four, measurement. Ask ChatGPT, Perplexity and Google's AI features the ten questions your buyers ask, record which companies are cited, and repeat monthly. Look at which of your pages appear as citations and which do not. Then keep going: one new independent mention per month beats any technical trick. The digital marketing service page describes how we run this as a monthly program alongside conventional SEO.

What not to do

Do not stuff pages with keyword variations. Answer engines extract meaning from clear sentences, and a page written for a keyword density target reads as noise to a model and to a buyer. Do not buy or fabricate reviews; platforms detect them, buyers detect them, and a model citing a review page with obvious fakes will cite the fakes. Do not create a dozen thin directory profiles with different descriptions, which makes the entity less consistent, not more.

Do not block AI crawlers because a plugin defaulted to it, and do not, on the other hand, expect a robots change alone to produce citations. And do not hire anyone who guarantees placement in AI answers. Nobody sells those positions. What a serious partner can guarantee is the work: consistent entity data, correct structure, a corroboration plan and honest monthly reporting. The same test applies here as when you choose a software development partner: ask what they will do, not what they promise.

The next step

Run the week-four measurement today, before changing anything, so you have a baseline. Then start with consistency, because it is free and it is the step everything else depends on. If you would rather have the audit done for you, with the checklist above applied to your site and a corroboration plan for your category, tell us your ten buyer questions and we will show you what the answer engines currently return for them.