How to improve your AI search visibility

Being recommended by an AI engine is not luck and it is not a trick. It is five pieces of work, done in a specific order, because each one is wasted if the one before it is broken.

Step 0: measure, before you change anything

Almost every business that starts this work guesses at the problem and fixes the wrong layer. There are only three possible problems, and they need completely different work:

• Engines cannot reach you. A technical problem. Fixable this week.
• They can reach you but cannot use you. A content and structure problem. Fixable this month.
• They can use you but do not trust you enough to name you. An authority problem. Months.

You cannot tell which one you have by looking at your website. Run a scan first, get the answer, then start at the step it points to.

Step 1: let the crawlers in

This is the most common hard failure and the most embarrassing, because the site owner almost never knows. Bot protection, a firewall rule or a managed robots.txt quietly returns a 403 to every AI crawler, and the business simply does not exist to those engines.

We know how easy it is to miss. The first time we scanned our own site we found our Cloudflare settings were blocking ClaudeBot, GPTBot and PerplexityBot. We were an AI visibility company that AI could not read.

Check whether the main crawlers get a 200 from your server: GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot and Google-Extended. Then check your robots.txt is your own file and not a managed one your host substituted. If you use Cloudflare, look specifically at Bot Fight Mode, the block-AI-bots setting and AI Crawl Control, because any of the three will do it silently.

One more in this category: if your content only appears after JavaScript runs, an engine that does not execute your scripts sees an empty page. View your own site with JavaScript disabled and see what is left.

Step 2: make your pages quotable

An answer engine is not ranking you, it is quoting you. It needs a passage it can lift cleanly and stand behind. Most marketing copy is unquotable: vague, hedged, and built to sound impressive rather than to be extracted.

Three changes do most of the work:

• Ask the question your buyer asks, as a heading, in their words. Not "Our Approach" but "How much does an AI visibility audit cost?". An engine matching a question to a page finds the page that asks it.
• Answer it immediately underneath, in 40 to 60 words, self-contained. No preamble, no "as we mentioned above". Something an engine can take whole without needing the rest of the page.
• Add statistics and quotations from named sources. Research from Princeton, Georgia Tech and the Allen Institute for AI found that adding quotations from named sources lifted generative-engine visibility by up to 41%, and adding concrete statistics by up to 30%. This page follows its own advice, which is the honest way to cite it.

Be specific about who and where. "We serve businesses nationwide" is invisible. "We fit boilers in Leeds and Bradford" is retrievable, because it matches how people actually ask.

Step 3: tell engines exactly who you are

Engines have to work out whether the business on your page is the same one mentioned somewhere else. If they cannot resolve that confidently, they leave you out rather than risk being wrong. Structured data removes the guesswork.

• Organization schema with your real name, URL and logo, plus sameAs links to your profiles elsewhere. That is the line that connects the entity together.
• FAQPage schema on any page that answers questions, so the questions and answers are machine-readable rather than inferred.
• An llms.txt file at the root, a plain-text summary of who you are and what your key pages are. It is an emerging convention rather than a guaranteed standard, and it costs an hour.
• A published date that moves when the page changes. A dateModified stuck two years in the past tells an engine the page is stale.

If another business has a name close to yours, this step is the one that separates you. Say so explicitly in your schema rather than hoping the engine works it out.

Step 4: earn the mentions engines actually cite

This is the step that moves the number, and the reason the other four are not enough on their own. Around 85% of the sources AI engines cite are third-party pages, not the brand's own website. You can do everything above perfectly and still not be named, because nothing outside your own domain says you exist.

What works, roughly in order of effort:

• Get into the roundups engines already quote. Find the "best X" and comparison pages that come up when you ask an engine your own buyer questions. Those specific pages are the ones being cited. Ask their authors to consider you, briefly and without a template.
• Answer real questions on Reddit. Reddit is consistently the single most-cited domain in AI answers. It is also the fastest way to get banned if you treat it as a marketing channel. Participate first, answer properly, and link only where the link genuinely helps the person asking.
• Get listed where your category is listed. Directories, industry bodies, Wikidata, Crunchbase. Unglamorous and durable.
• Publish something worth citing. Original data beats opinion. If you have numbers nobody else has, that is the asset other people link to.

Mentions matter even without a link. An engine reading your brand name next to your category on a page it trusts is the signal, whether or not there is an anchor tag around it.

Step 5: treat each engine separately

ChatGPT and Perplexity overlap only around 11% in the domains they cite. Treating them as one audience means most of your work lands on one engine and misses the other.

Start with Perplexity. It re-crawls quickly and shows its sources openly, so it gives you a visible win early, which matters when the rest of the work takes months. Then work outward. The practical method is simple: ask each engine your six most important buyer questions, write down who it names, and re-run it monthly. That is your scoreboard.

What to expect, and when

Being straight about this is the difference between the work paying off and you concluding it failed.

• Days to weeks: technical fixes. Crawlers start getting through, pages get indexed, and you can see the change in your server logs before it shows in any score.
• Weeks: structure and content work. Engines re-crawl, and your pages become usable to them.
• Months: being named. This is the one everyone wants first and it arrives last, because it depends on other people publishing things.

Watch the leading indicators rather than the headline number, or you will lose heart in week three. AI crawler hits appearing in your logs, pages moving into the index, and your name showing up in an engine's cited sources before it recommends you: all three come before the score moves.

We are running this exact process on ourselves in public, including the parts that have not worked yet. The current number is on our proof page, updated daily, whatever it says.

Questions, answered

For the same material as a ranked practice list rather than a method, see AEO best practices in priority order.