Nobody outside Google, OpenAI or Anthropic can see exactly how their systems choose which sources to cite. That will not change much soon. What you can control is whether those systems can actually read your website, pull a clean answer out of it, and confirm the facts. For most B2B sites, fixing that is the highest-return AI search work available today.
Regulators are pushing for transparency
Slobodan Manic makes this case in a Search Engine Journal article that starts with regulation. According to the piece, on June 17 the UK Competition and Markets Authority used Google's Strategic Market Status designation to impose two binding requirements.
- Google must rank organic results, including inside AI Overviews, by objective and non-discriminatory criteria.
- Google must give businesses more transparency on how ranking works, advance notice of major ranking changes, and a documented way to complain.
That is real recourse for publishers and businesses. But Manic's central point is that even full transparency would not change what machines need from a page. Knowing the rules settles arguments. It does not do the work.
Related: Unlocking B2B Growth: A Practical Checklist for Leveraging Intent Data (2026)
Why opacity is the wrong thing to worry about
Many teams wait for clarity before investing in AI search. They want to know the ranking factors, the citation logic, the weighting. That wait is costly, because the basics every AI system relies on are already well understood: it has to fetch your page, parse the content, find a passage that answers the question, and trust that the passage is accurate.
If any of those steps fail, no amount of clever optimisation helps. And those failures are common on B2B sites built for design awards rather than for machines.
Check 1: Can machines see your content?
Many modern sites render content with JavaScript. Browsers handle it fine. Crawlers and AI fetchers may not run scripts at all, or may run them only partially. If your product description, pricing or feature list only appears after JavaScript runs, some systems will see an empty page.
- Open your key pages with JavaScript disabled in your browser.
- Check that headings, body copy, pricing and key facts are all still visible.
- View the page source and search for a sentence from your main content.
- If it is missing, work with developers on server-side rendering or static generation for important pages.
This check matters most for pages built with single-page app frameworks, pricing calculators and interactive product tours. Those are often the pages buyers care about most, and the ones machines are most likely to see as blank. You do not need to rebuild the whole site. Rendering the key facts in plain HTML on your top commercial pages is usually enough to start.
Related: Buying B2B Data: Complete Checklist Before You Purchase (2026)
Check 2: Is the answer easy to extract?
AI systems cite passages, not whole pages. If the answer to "who is this for?" or "how much does it cost?" is buried inside a long story, the model has to work harder to find it, and may pick a competitor's page that states it directly.
- Use descriptive headings that match real questions buyers ask.
- Put the direct answer in the first sentence under each heading, then add detail.
- Keep one topic per section so passages stand on their own.
- Use lists and tables for comparisons, specifications and steps.
- Avoid hiding important facts in images, carousels or accordions that load late.
A simple test: read only the first sentence under each heading on a page. If those sentences alone answer the main questions a buyer would ask, the page is in good shape for extraction. If they are all scene-setting, rewrite them.
Check 3: Can the facts be verified?
Models are more likely to trust a claim that is stated plainly and consistently. Your company name, what you do, who you serve, where you operate and what you charge should read the same way on your homepage, product pages, pricing page, About page and external profiles.
Contradictions cause problems. If one page says you serve "all businesses" and another says "mid-market SaaS," a model cannot be sure. For example, Happierleads states on its site that it identifies the exact person and company behind a website visit, and that EU and EEA visitors are identified at company level. Saying it the same way on every page, including pricing, gives machines one clear fact to repeat.
A 30-minute audit for your top pages
Pick your five most commercially important pages, usually the homepage, main product page, pricing, a key use case and your About page. For each one, record:
- Whether core content appears with JavaScript off.
- Whether a stranger could find a one-sentence answer to "what is this and who is it for?" in under ten seconds.
- Whether the facts match the other four pages exactly.
- Which claims have no supporting detail or source.
Fix the worst page first. Most teams find at least one important page that machines cannot read properly.
Measure the traffic, not just the citations
AI visibility tools can tell you when you are cited. They cannot tell you who came to your site afterwards. Buyers who discover you through an assistant often land on your site directly and browse without converting. Website visitor identification shows which companies and people are behind those visits, so you can tie machine-readable improvements to real pipeline.
The takeaway
You do not need to understand AI ranking to improve your odds. Make sure your key pages render without JavaScript, state answers in clean, extractable passages, and repeat the same verifiable facts everywhere. Run the 30-minute audit on your top five pages this week. That work pays off whether ranking systems stay opaque or regulators force them open.
