For twenty years, being found online meant one thing. Rank on the first page of Google, earn the click, make your case on your own site. That model is not dead, but it is no longer the whole story. A growing share of buying research now happens inside a chat window, where the customer never sees a results page at all. They ask ChatGPT which tool to use, or ask Perplexity to compare three vendors, or type a question into Google and read the AI answer at the top. The assistant answers in prose. It names a handful of companies. Then the conversation moves on.

If your business is one of the names, you just entered a shortlist you did nothing to join that day. If it is not, you were invisible in a way that no analytics dashboard will ever show you. There is no impression to count and no ranking to check. The customer simply never learned you exist.

We have spent the last few years doing this work for clients and for digital properties we operate ourselves, and the honest summary is this. Getting cited by AI assistants is not magic and it is not luck. It is deliberate, checkable work, and most businesses have not started it.

What actually changed

The old funnel had a moment of choice in the middle. Ten blue links, and the user picked one. Your job was to be in that list and to win the click. The new funnel removes the choice. The assistant reads a set of sources, synthesizes one answer, and presents it as settled. The user can expand the citations, and some do, but the framing has already happened. The assistant decided which companies were worth mentioning before the user saw anything.

That shifts the competition. You are no longer competing for a click against nine other results. You are competing to be in the source set the model draws from, and then to be the kind of source a model quotes rather than skims past. Those are different problems than classic ranking, and they reward different work.

One more change matters. Assistants answer specific questions, not keywords. Nobody asks an AI "project management software." They ask "what should a ten-person agency use to track client work, and why." The businesses that get cited are the ones whose pages actually answer questions like that, plainly, with a defensible claim near the top.

How citation actually works

Strip away the mystique and the mechanics are legible. When an assistant answers a question about companies, products, or services, it is usually doing some version of retrieval. It runs searches, pulls a set of pages, and assembles an answer from what those pages say, citing the ones it leaned on. The model's training data matters for background knowledge, but for anything current or commercial, the retrieved sources carry the answer.

So the practical question becomes, what gets a page into that retrieved set, and what gets it quoted once it is there.

The first gate is retrievability. Different systems build their source sets differently. Some lean heavily on the top results of classic search engines, which means your existing search visibility is doing double duty whether you planned it or not. Others cast a wider net and pull from forums, directories, comparison pages, and documentation that never cracked page one. This is why we treat SEO as the foundation, not the competition. A site that ranks nowhere has a much harder path into any AI answer, and a site that ranks well has already cleared the tallest gate for several of the major assistants.

The second gate is extractability. A model working through a retrieved page is looking for clean, liftable claims. A clear statement of what the company is and who it serves. A direct answer to the question, stated early, not buried under eight hundred words of throat-clearing. Headings that match the questions people actually ask. Structured data that confirms in machine-readable form what the prose says in English. Pages built this way get quoted. Pages built as mood boards with paragraphs get skipped, even when they were retrieved.

The third gate is trust, and it lives mostly off your own site. Models cross-reference. If your site says one thing, your LinkedIn says another, and third-party mentions say a third, you read as noisy, and noisy sources get discounted. If independent pages describe you consistently, in comparison articles, in directories, in discussions where practitioners talk shop, you read as an established entity, and established entities get named. Much of the highest-value work in this discipline happens on pages you do not control, earned the slow way.

There is a name emerging for all of this, Answer Engine Optimization, or AEO. The label will probably keep shifting. The work underneath it will not.

What a business should actually do

The to-do list is concrete. Here is where we start, in roughly this order.

Establish a baseline. Ask the major assistants the questions your customers ask. Who does what you do. What are the best options in your category. Is your company legitimate. Record what comes back, which competitors get named, and which sources get cited. This takes an afternoon and most companies have never done it. You cannot improve an answer you have never read.

Fix your entity story. One canonical description of what your company is, used verbatim on your site, your LinkedIn, your registry listings, and every profile you control. Organization schema on the site that matches it. Machines cross-check these sources, and consistency is the cheapest trust signal available.

Answer real questions on real pages. Take the questions from your baseline and make sure a page on your site answers each one directly, with the answer in the first screen, a heading that mirrors the question, and specifics a model can lift. This is also just good writing for humans, which is not a coincidence. The overlap between what an assistant will quote and what a busy person will read is nearly total.

Keep the search foundation solid. Crawlable site, fast pages, clean titles, an honest sitemap, and a robots policy that admits the AI crawlers rather than blocking them by default. Being reachable by GPTBot and PerplexityBot is table stakes, and plenty of sites fail it by accident.

Earn third-party presence. Get described accurately in the places assistants retrieve from. Comparison pages, industry directories, communities where your category gets discussed. This is slower than everything above and it compounds harder than any of it.

Re-measure on a schedule. Rankings update daily and dashboards make them easy to watch. AI answers have no dashboard yet, so you build your own habit. Same questions, same assistants, monthly. Watch which sources appear, and aim your next quarter of work at them.

The quiet advantage

Here is the part worth acting on. Most of your competitors are not doing any of this. They are watching the same rankings they watched in 2019, while a growing slice of their market asks questions they will never appear in. The window where deliberate work beats incumbency does not stay open forever. In classic search it closed years ago, and the businesses that moved early are still collecting on it.

When someone asks an AI who to trust, the answer comes from somewhere. It comes from sources that were retrievable, quotable, and consistent long before the question was asked. That is buildable. Start with the baseline. Read what the machines currently say about you. Then close the gap, one checkable piece at a time.