You can't optimise a website an AI can't read
Most AI visibility tactics fail because they sit on a website the AI assistant cannot actually read. Bolt-on fixes such as llms.txt files and FAQ schema assume the AI already reaches, reads and believes your site; on a slow, bloated, script-heavy build it does none of the three. AI search visibility is a foundation problem before it is a tactics problem: fix speed, legibility and consistency first, and the tactics start to stick.
The wall the agent won't climb
When an AI assistant researches on a buyer's behalf, it gives your website a few seconds and no second chances.
Picture the visit. A buyer asks an assistant to compare firms in your category. The assistant fetches your homepage while it writes its reply. What greets it is what greets every visitor to a typical business site: megabytes of JavaScript, a cookie wall, a chat widget, a tag manager loading a dozen trackers, and the actual content - the part that says what you do and why you are good at it - assembled by scripts the assistant may never run.
A human visitor might sigh and wait. The assistant does neither. It has an answer to write and a budget of seconds to write it in, so it takes whatever plain text your page surrendered and moves to the next source. It doesn't complain, retry or warn you. It builds the shortlist out of the sites that were easy to read, and yours wasn't one of them.
Your content might be brilliant. Behind that wall, it was never read.
Tactics on a broken foundation
Most AI visibility tactics assume the assistant is already inside, reading comfortably. On many business websites it never gets that far.
You will recognise the shopping list, because someone has probably pitched you part of it: an llms.txt file, FAQ schema, "AI-optimised" blog posts, a sprinkle of keywords the robots supposedly favour. To be fair to the list, most of these have a place - on a site that works. We ship several of them as standard. But consider what the data says about the most fashionable item: 97% of llms.txt files received zero requests from AI bots (Ahrefs, 2026). A file nobody reads is not a fix. It is a receipt.
The same logic runs through the rest. Schema that claims what your page cannot visibly show gets treated as spam. Answer-shaped content that only exists after JavaScript executes might as well not exist. Every tactic inherits the site underneath it.
Bolting them onto a slow, bloated build is lipstick on a wall the agent won't climb. The spend does not fail loudly; it just leaks, month after month, while the report says "implemented".
What "foundation" actually means
The foundation is everything that decides whether an AI can reach, read and believe your site before any tactic applies. It is concrete and checkable:
- Fast to fetch. Google's bar for a good page is rendering in under 2.5 seconds (Google, 2024). An assistant assembling an answer in real time is less patient than that. Speed is the first trust signal, and the first filter.
- Legible without JavaScript. What a plain fetch of your page returns is what the AI gets. If your value proposition only appears after scripts run, to an answer engine it does not exist.
- Lean by construction. Every plugin, tracker and framework adds weight to the read and risk to the business - 91% of WordPress vulnerabilities live in plugins (Patchstack, 2025). The cheapest page to read is the easiest to cite.
- Consistent in its facts. Your name, offer and story should read identically on your site and everywhere else you appear. Muddle is the gap the AI fills by guessing, and it rarely guesses in your favour.
There are no tricks on that list. The foundation isn't a tactic; it's the surface tactics need to grip.
The three things an AI needs before it recommends you
Before an AI assistant stakes its credibility on your business, it must be able to do three things, in order. Each one depends on the last.
- Reach you. Your site responds quickly, and nothing - a misconfigured robots.txt, an overzealous "block AI bots" switch at your host - turns the assistant away at the door.
- Read you. Your meaning sits in the HTML in plain language: who you are, what you do, for whom, at what price. Structure and answer-first writing do the heavy lifting here.
- Believe you. A recommendation is a transfer of trust. The assistant cross-checks you against the rest of the web, and names you only if the story holds together everywhere it looks.
The arithmetic is uncomfortable. Almost everything sold as "AI SEO" operates at stages two and three; if stage one fails, the rest is theatre. And stage one fails quietly, which is why so many businesses are paying for optimisation the AI never sees.
Introducing AI Search Foundations
AI Search Foundations are the base layer of AI search visibility: a website fast enough to be fetched in the moment an AI answers, legible enough to be read without executing scripts, consistent enough in its facts to be believed, and minimal enough to be trusted. Tactics sit on top of this layer, and they inherit its strength or its weakness.
Naming the layer changes the question you should ask any provider, including us. Not "which AI tactics do you offer?" but "will your tactics stick to the site I actually have?" An honest answer sometimes hurts: if the foundation is broken, the right first spend is fixing it, not decorating it. It is why we build the foundation before we touch a single tactic - optimisation applied to a site the AI can already read is the only kind that compounds.
The full standard - found, understood, trusted, secure, converting - is laid out in the AI Search Foundations checklist, and you can score your own site against it right now. Our free diagnostic reads your website the way an AI assistant would and grades the foundation across 40+ checks, in about ten seconds, with no email required.
You can't optimise a website an AI can't read. So before anyone sells you the lipstick, find out whether the wall holds.
Frequently asked questions
Why is my site not showing up in AI answers?
Start with the unglamorous explanation: the AI may not be able to read your site at all. If your pages load slowly or your content only appears after JavaScript runs, an AI assistant fetching sources in real time gets little or nothing back, and quietly uses a competitor it can read. Legibility comes before authority.
I paid for AI SEO and nothing changed. Why?
Most likely the tactics were sound and the foundation was not. Content tweaks, schema and llms.txt files all assume the AI can already reach and read your site quickly. If it cannot, the tactics never get seen. Score the foundation first, fix what blocks the read, then the same tactics start paying.
What are AI Search Foundations?
AI Search Foundations are the base layer of AI search visibility: a website fast enough to be fetched in the moment an AI answers, legible enough to be read without executing scripts, consistent enough in its facts to be believed, and minimal enough to be trusted. Tactics sit on top of this layer and inherit its strength or its weakness.
Do I need to rebuild my website?
Not always, but be honest about the maths. Some sites need a handful of fixes; others carry so much accumulated weight - plugins, trackers, page builders, a database serving a brochure - that patching costs more than rebuilding lean. A foundation score tells you which side of that line you are on before you spend.
Is llms.txt worth adding?
Yes, as cheap insurance, and no, not as a strategy. It costs nothing and may pay off as adoption grows, which is why we ship one as standard. But 97% of llms.txt files got zero AI-bot requests (Ahrefs, 2026), so never mistake the file for the fix. The read still happens on your actual pages.