The Search Box Moved
For twenty years the deal was simple: a customer typed a question into Google, Google showed ten links, and the website that ranked got the visit. In 2026 that deal is being renegotiated. ChatGPT, Claude, Perplexity, Gemini, and Google's own AI Overviews and AI Mode now sit between a growing share of buyers and the websites they would once have visited. The customer asks, the assistant answers, and the answer either mentions you or it doesn't.
The question for a business is no longer only "do we rank for this keyword." It is "when someone asks an assistant who does this in our category, are we part of the answer, and does the assistant get our details right." If the model can't read your site, can't tell what you do, or can't verify who you are, the answer is no, no matter how good the work behind the site is.
This is not a replacement for SEO. It is an extension of it. Some people call it answer engine optimisation, some call it generative engine optimisation, some just call it AI visibility. The label matters less than the mechanics, and the mechanics are surprisingly concrete.
How an LLM Actually Finds You
There are three ways your content reaches a language model. The first is training data: what the model memorised when it was built. That is slow, out of your control, and increasingly less important for business questions. The second is live retrieval: when a user asks a question, the assistant runs a web search, fetches a handful of pages in real time, reads them, and composes an answer with citations. ChatGPT search, Perplexity, Claude with web access, and Google's AI Mode all work this way. The third is direct crawling by AI-specific bots, such as GPTBot, ClaudeBot, PerplexityBot, and Google-Extended, which feed both of the others.
For a business website, live retrieval is where the action is. It means your page gets fetched, parsed, and quoted at the moment someone is deciding. And the signals that decide whether that happens are the same ones that have always driven search: can the page be crawled, is it clear what it says, and is there reason to trust it. The difference is that a language model is far less forgiving of ambiguity than a ranking algorithm was.
What the fetcher sees is not what you see
Most AI fetchers do not execute JavaScript, or do so inconsistently and with tight timeouts. A website that renders its content in the browser, after the page loads, can look like an empty shell to the machine: a header, a footer, and a loading spinner. Everything your designer agonised over is simply not there. Server-rendered HTML with the real text in it is the baseline, and a surprising number of sites built in the last five years fail it.
Seven Things an AI-Ready Website Gets Right
1. The content is in the HTML, not behind a script
Open your page, view the source, and search for a sentence from the middle of the content. If it is not there, a language model probably can't see it either. Modern frameworks can render on the server by default; the fix is usually a build decision, not a rewrite. For pages that must be interactive, make sure the words are delivered first and the interactivity layered on top.
2. The structure reads like an outline
Models break pages into chunks and reason about each chunk on its own. One clear heading per page, section headings that state the point rather than tease it, short paragraphs, and lists or tables for anything comparable: prices, specs, locations, steps. A heading that says "Our Services" tells a model nothing. A heading that says "Web development for Austrian and German SMEs" tells it exactly what the section is about.
3. The content is shaped like an answer
Lead with the answer, then elaborate. Who you are, what you do, where you do it, for whom, and what it costs, stated plainly in the first lines rather than buried under a brand manifesto. FAQ sections work unusually well because they map directly to how questions are asked. Specific, verifiable facts get cited; vague claims of excellence get skipped.
4. Structured data restates the page for machines
JSON-LD is a machine-readable restatement of what the page already says: Organization or LocalBusiness for who you are, Article with author and dates for what you publish, Product or Service for what you sell, FAQPage for your questions, BreadcrumbList for where the page sits. It resolves ambiguity a model would otherwise have to guess at. There are many companies with similar names; structured data is how yours gets attached to the right facts.
5. Your entity is consistent everywhere
Assistants cross-reference. The name, address, description, and founder on your site should match what is on LinkedIn, your Google Business Profile, the directories, and the press mentions. Inconsistency reads as uncertainty, and an uncertain model hedges or omits. One canonical description of the business, used everywhere, is one of the cheapest wins available.
6. The door is deliberately open to the right bots
Your robots.txt should make a conscious decision about which AI crawlers are allowed, rather than inheriting whatever a plugin set years ago. Your sitemap should be current and honest. An llms.txt file, a plain Markdown index of the pages that matter and what they contain, is an emerging convention with uneven adoption. It is not a guarantee of anything, but it is cheap, harmless, and signals intent.
7. Nothing stands between the fetcher and the text
Fetchers have timeouts and no patience. Cookie walls that hide content until accepted, interstitials, login gates on pages that should be public, and multi-second load times all end the same way: the model moves on to the next result. Speed and clean markup are not only a Core Web Vitals concern anymore; they decide whether you are read at all.
What to Keep Away From the Machines
Open does not mean everything. Negotiated pricing, client data, internal documentation, drafts, and anything you would not want quoted out of context should stay behind a deliberate wall. The point is that it should be a choice you make, page by page, rather than a default you inherited. A site that opens the right pages and closes the right pages is in a stronger position than one that does either blindly.
Measuring Whether It Works
There is no single dashboard for this yet, but it is measurable. Ask the assistants directly: take the ten questions your buyers actually ask and run them through ChatGPT, Perplexity, Claude, and Google's AI Mode once a month, and note who gets cited. Check your server logs for AI user agents to confirm you are being fetched at all. Watch for referral traffic from the assistant domains in your analytics. Trend it over time, the same way you would track rankings.
Where to Start
Begin with an honest audit: fetch your own pages the way a bot would, with JavaScript disabled, and see what survives. Then fix in order of leverage: rendering first, because nothing else matters if the text is invisible; then structure; then structured data; then entity consistency; then bot policy. Most sites built on a modern framework in the last few years can be made AI-ready without a redesign. Sites that are a wall of client-rendered components, or a page-builder stack of nested boxes, usually cannot, and that is worth knowing before the next marketing budget is set.
The website that wins in 2026 is the one a machine can summarise in a paragraph and a human can trust at a glance.
This is how we build at UnityWorld: server-rendered pages with the content in the HTML, structured data on every page, one consistent entity across three languages with proper alternates, and a bot policy chosen on purpose. None of it is exotic. It is simply the difference between a website that was built for a browser and one that was built for the way people actually find things now.
