How to See What Your Business Really Needs in AI Search.

Most businesses decide they need help with AI search the week a competitor gets named by ChatGPT and they don't. The usual reaction is to buy a tool or hire someone to "do GEO", which skips the step that would tell them what is broken. You can't fix how you show up in AI answers until you've seen how those systems read your business today, and almost nobody has looked.

Looking is cheap, though. A handful of checks, most of them free, will show you whether AI assistants can even reach your content, whether they trust you enough to cite you, and whether they understand what you sell. Whatever comes back from those checks is your real to-do list, and it tends to be shorter and more specific than the blanket "invest in AI search" advice everyone is repeating. This is the diagnosis I run for a business before I recommend changing a single thing.

Why your old search scoreboard is going dark

The rankings-and-clicks dashboard business owners have watched for years is measuring less every quarter, and that is the first thing to understand before you spend anything. Pew Research followed the browsing of 900 US adults and found that when Google showed an AI summary, people clicked a normal search result in only 8 percent of visits, against 15 percent when there was no summary, and the source link inside the summary itself got clicked in just 1 percent of visits. Rankings still move, but a shrinking share of them turn into someone actually landing on your site. So if you judge your AI search health by where you rank, you're reading a gauge that is slowly disconnecting from the outcome you care about. A page can sit at position three, get folded into an AI answer, and send you almost nothing, while a business you outrank gets named as the source instead. The diagnosis has to stop treating the ranking report as a stand-in for visibility and start looking at what the assistants say and cite, which is why I've argued before that keyword ranking is a weak KPI on its own.

Start by asking the assistants about your own business

The fastest diagnostic costs nothing and takes about ten minutes. Open ChatGPT, Gemini, and Perplexity, and ask each of them the questions your buyers would ask before choosing a company like yours. Ask the category question rather than your brand name. Something like which companies handle X in your city, or what the best option is for Y for a business your size. Then read the answers the way a prospect would read them, with a cold eye.

You're checking three things in what comes back. Whether you appear at all, whether the assistant describes you accurately, and who it names in your place or beside you. A business that never surfaces has a discovery or trust problem, while one that surfaces with the wrong description, an old service line, or a price you changed a year ago is dealing with something on the content and entity side instead. And when the same two or three competitors keep getting named across all three assistants, that is your real competitive set in AI search, which is often not the set you assumed you were fighting.

Run the plain brand-name query too, the who is your company version. If the assistant hedges, blends you with a similarly named business, or gets your specialty wrong, you've found something to fix before any of the more advanced work. I went through what a strong answer and a weak answer look like in the piece on owning your AI search visibility instead of renting it, and the difference is usually easy to spot once you know what you're reading for.

See what AI crawlers actually read on your site

Plenty of businesses turn out to be invisible to AI for a boring but common reason, which is that the crawlers can't read the page in the first place. Vercel analyzed the major AI crawlers and found that none of them render JavaScript. GPTBot, ClaudeBot, PerplexityBot, and OpenAI's search crawler fetch your HTML and stop there.

If your services, your prices, or your key descriptions get drawn onto the page by JavaScript after it loads, those crawlers see an empty shell where your selling points should be. Your browser hides this from you, because it runs the JavaScript and paints the finished page, so everything looks perfectly fine when you check it yourself.

The same study found the AI crawlers waste about a third of their fetches on pages that return errors, so a page that hands bots a broken response while showing visitors something normal can disappear without you ever noticing. That is why the check has to look at the raw response rather than the rendered page, and it's the part most quick audits leave out.

We built a free Chrome extension called Agent View for exactly this gap. It compares what a normal visitor sees against what an AI crawler receives, re-fetches your pages as GPTBot and ClaudeBot to catch the ones that return an error only to bots, checks your robots.txt and llms.txt, and outputs a prioritized fix list. Run it on your homepage and your top service pages, and if the AI view is missing big chunks of what a customer would see, you've found the highest-priority fix on your list, because none of the other work counts for anything while the systems can't read you.

Check whether AI treats your site as a source

Showing up in an AI answer and being the source it credits are different outcomes, and the gap is wider than most people expect. Ahrefs studied a million keywords that trigger Google's AI Overviews and found that a page ranking number one organically appears among the top three cited sources only about half the time. Ranking well helps, but it doesn't hand you the citation, so you can be the best-ranked page in your category and still watch the AI credit someone else for the answer.

For the diagnosis, keep two questions apart. Does the assistant mention your brand in its answer, and does it point to your site as where the information came from. A mention builds familiarity over time, while a citation is what sends a reader to you and tells the model your page is a primary source it can lean on. If you're getting mentioned without being cited, the work ahead is about becoming the clearest and most complete answer to that question rather than settling for a name-check.

The difference shows up in a single query. When an assistant recommends vendors in your category, two kinds of names come back. Some brands get listed because the model has absorbed them from being written about across the web, and some get an actual link to their own page because that page answered the question directly and thoroughly. The first kind is coasting on reputation and can be displaced the moment a rival publishes something better, whereas the second has earned its place on the page and tends to hold it. Knowing which one you are changes what you build next, and a colleague of mine wrote a detailed breakdown of what earns the citation when backlinks no longer decide it that is the natural next read once you've found your gap.

Map the questions your buyers actually ask

AI search rewards businesses that cover the full set of questions around what they sell, well beyond the single keyword they used to chase. A prospect deciding on a company like yours works through a chain of questions over days, from what this costs to how it differs from the alternative they're also weighing, whether it's worth it at their size, and who is actually good at it. If all you have is a homepage and a services page, you've answered almost none of that, and the assistant has nothing of yours to reach for when those questions come up in a conversation.

So write the chain down. List every question a buyer moves through between first hearing about your category and signing a contract, then mark which ones you've answered somewhere a crawler can read. The gaps tend to jump out once they're on paper, and they point at specific pages to build rather than a vague instruction to make more content. This mapping is the core of the Topic Ownership Strategy we run with clients, where one pillar topic and the cluster of buying questions around it get covered on purpose instead of one blog post at a time.

A quick example makes it concrete. Say you sell accounting software to small Indonesian retailers at around Rp 300 thousand a month. The buyer's chain runs from what accounting software costs for a small shop, through whether it handles PPN and local tax reporting, whether it beats a spreadsheet, how it compares to the two names their friend mentioned, and finally whether anyone their size actually uses it. That's five distinct pages of buying intent, and most software sites answer the first one and none of the rest. When an assistant fields any of those later questions, it pulls from whoever did answer them, which is how a smaller competitor with better content ends up recommended over a bigger brand with a thinner site.

Make sure AI knows who your business is

Behind every AI answer about a company sits the model's sense of what that company is, and that sense is assembled from signals you can influence, from your schema markup to the way your business is described consistently across your own pages and the links between your official site and your profiles elsewhere. When those signals are thin or contradictory, the model gets unsure about your identity, which is when it confuses you with a similarly named company or describes you with a service you dropped two years ago. You don't need to read code to sense-check most of this, since it comes down to whether your site tells one consistent story about who you are and what you do, and whether your name, your focus, and your location match across your website, your Google Business Profile, and your LinkedIn rather than drifting apart. Google's own structured data documentation covers the markup side if you want to hand your developer something concrete. When we ran this check on our own business we found gaps in our entity schema and missing links between the founder profile and the company, and we build this for a living, so most sites carry more of these gaps than their owners would guess. Closing them is some of the cheapest trust you can earn in AI search, because you're not creating anything new, you're making the story you already tell legible to a machine.

Turn what you found into a short priority list

By the end of those checks you're no longer guessing about AI search, you have a specific list, and the order to work through it almost writes itself. Crawlability comes first, because a site the systems can't read gets nothing from any other fix. Entity and schema signals come next, since they're cheap and they decide whether the model is confident enough to use you at all. Then the content for the buying questions you can't currently answer, which is the slower, compounding work that turns visibility into real pipeline over a few quarters.

One thing to plan for is that this is not a one-time check. AI answers shift as models get retrained and as competitors publish, so the same query that ignored you in March can name you in June, or drop you again by September. Re-running the assistant questions and the crawlability check once a quarter is enough to catch a slide early, and it turns the diagnosis into a habit rather than a panic every time you notice a competitor getting recommended.

What you'll notice across most of these lists is how little budget they actually need. The businesses I look at usually need three or four targeted fixes rather than a platform subscription and a retainer for content nobody asked for. If you'd rather not run the diagnosis yourself, our free AI Search Visibility Audit does exactly these checks on your site and hands you the prioritized list, and you can see how we track the results over time on our measurement page before you commit to anything.