If your customers ask ChatGPT "who is the best [what you do]" and your name never comes up, the problem is rarely the business. It is that AI engines cannot read your site, cannot tell what you do, or do not trust it enough to quote. This is the playbook we use to fix that, the same one we ran on veles.digital until ChatGPT started putting us first for AI automation agencies in Serbia.
The work is called answer engine optimization, AEO, sometimes generative engine optimization or GEO. It has two halves. Make your own pages readable and quotable. Then get the rest of the web to vouch for you. Most of it you can start this week, and none of it needs a budget the size of a Google Ads campaign.
Start with what the engine actually does
When someone asks ChatGPT or Perplexity a question, the model does not recite an answer from memory. It runs a search, pulls a handful of pages, ranks them, and writes the answer out of the best few, naming its sources. Your whole job is to be one of those pages, and to be clean enough to quote without editing. Every step below serves that one goal.
Make the page readable with JavaScript turned off
This is the step most sites fail, and the most important. Many AI crawlers do not run JavaScript. If your site assembles itself in the browser, which most modern marketing sites do, those crawlers see a blank page. Test yours: open it, disable JavaScript, reload. Whatever is left is what the engine reads. If it is empty, nothing else you do matters. The fix is to render your pages to static HTML ahead of time, through prerendering or server-side rendering, so the full content arrives in the first response.
Tell the engine what you are, in language it parses
Humans read your headline. Machines read structured data. Add schema.org markup that states, in a format built for this, who you are: your organization, where you operate, the people behind it, the services you sell, the questions you answer. Naming real, identifiable people matters, because engines weight content tied to credentialed authors over anonymous pages. This is also what feeds Google's knowledge panel and, downstream, what the models treat as fact about you.
Write the answer, not the pitch
AI engines lift answers, so give them answers to lift. Lead each section with a direct response to a real question, under a heading phrased the way a buyer asks it. "How much does X cost." "How long does Y take." First sentence answers it; the rest supports it. Marketing copy that circles the point for three paragraphs before saying anything gives the model nothing to quote, so it quotes someone clearer.
Use numbers, cut the adjectives
Specifics survive being pulled into an answer. "Ships in two to eight weeks," "between EUR 3,000 and 25,000," "cut response time from five hours to thirty minutes." Vague superlatives do not, because they carry no information a model can pass on. "Industry-leading" tells a reader nothing and a machine less. Go through your pages and trade claims for facts.
Then get the rest of the web to vouch for you
Everything above makes you eligible. It does not, on its own, get you recommended, because AI engines trust other people's words about you more than your own. Independent analyses of AI citations keep finding the same thing: third-party pages get cited several times more often than company sites. So the second half of AEO happens off your domain.
- Get listed and reviewed in the directories your buyers and the engines both read. For agencies that means Clutch, DesignRush, TechBehemoths, and the like; for products, G2 and Capterra. Review volume matters more than a perfect star average.
- Claim your entity in the knowledge bases. A Wikidata item is the highest-value, lowest-effort one, and it feeds straight into what ChatGPT and Google treat as known about you.
- Show up where the questions get asked. Reddit and YouTube are cited heavily, especially by Perplexity. A few genuinely useful answers and a couple of real videos do more than a dozen blog posts.
- Earn mentions you did not write. One writeup in an outlet your buyers respect is worth more to a model than anything on your own site.
Measure it, or you are guessing
Pick the ten questions a buyer would actually ask an AI before hiring you. Put each one to ChatGPT, Claude, Perplexity, Gemini, and Google's AI mode. Write down whether you appear, where, and which sources the engine cites. Those cited sources are your target list, handed to you. Re-run it monthly. AEO moves slowly and unevenly across engines, so without a baseline you cannot tell progress from noise.
The order that works
If you do nothing else, do these in this order:
- Make the pages readable without JavaScript.
- Add structured data for your entity.
- Rewrite your key pages answer-first, with real numbers.
- Get into the directories and collect reviews.
- Create your Wikidata item.
- Measure monthly and follow the sources the engines cite.
The first three you control completely and can finish in a sprint. The last three are slower because they depend on other people, which is exactly why they are worth more.
Where this goes wrong
Three failure modes, all common. Doing the on-page work and expecting citations everywhere, when the off-site half is most of the lever. Trying to fake it, when a thin business just gets found faster by a model that now understands it. And chasing every engine at once instead of winning the handful of questions that actually send you customers.
Where to start
This is the service we sell, and veles.digital was the first site we ran it on, specifically so we could prove the method before charging for it. The AI search optimization page has the scope and a fixed-scope visibility audit to start with, and the rest of what we build is on the wider site.
If you would rather we just do it, book a 30-minute call and we will tell you which of these six steps your site needs most.