More questions end without a click every year, and the answer people read is written by a model that never runs your JavaScript. If your pages only exist after a script builds them in the browser, there is nothing for the model to read, let alone repeat. We build sites that arrive as plain readable text on the first request, then keep them worth quoting.
A person opens your site and a browser spends a second or two running scripts to assemble it. An AI crawler does not do that step. Vercel and MERJ tracked more than 500 million crawler requests and found no evidence of JavaScript being executed at all. Whatever your page looks like before the scripts run is the entire page, as far as the model is concerned.
Licensed, insured, and answering the phone at two in the morning. Serving homes and small commercial buildings for eleven years.
Book a visit<!doctype html> <html> <head> <title>Home</title> <script src="/bundle.js"></script> </head> <body> <div id="app"></div> </body> </html> No headline. No services. No hours. Nothing to quote, and nothing to cite.
<!doctype html> <html> <head> <title>Emergency repairs, same day</title> </head> <body> <h1>Emergency repairs, same day</h1> <p>Licensed, insured, answering the phone at two in the morning.</p> <h2>What we handle</h2> <ul><li>Burst pipes</li> … </body> </html>
Type view-source: in front of your web address and load it. If you can find your headline in that text, a crawler can too. If you mostly see script tags and an empty container, you have your answer, and no amount of content work fixes it from the outside.
A model writes its answer out of whatever text it could read. It will not wait for your scripts to finish, and it will not guess at what a page meant to say. Plain sentences that answer the question give it something to work with. An empty container sends it to a competitor who wrote plainly.
who can fix a burst pipe tonight
Harborline Plumbing takes emergency calls overnight and lists same day repair as a standard service.harborline.example
They are licensed and insured, and have been working on homes and small commercial buildings for eleven years.harborline.example
Common jobs listed include burst pipes, failed water heaters, and sewer backups.harborline.example
A local trade directory lists the same overnight availability, which matches what the company says about itself.directory.example
The page had to arrive as text on the first request. If the words only appear after a script builds them, none of the rest of this list matters.
The page said the thing plainly and early, in a sentence that still makes sense pulled out on its own. Bury it in paragraph nine and nothing pulls it.
Licensed, insured, eleven years, what they handle. A vague page gives a model nothing to repeat, so it gets summarized in a line and dropped.
Mentions off your own site carry the most weight and are the part nobody fully controls. Anyone promising to control it is selling you something they cannot deliver.
Search did not go anywhere, but more of it now ends on the results page or inside a chat window, and the click you used to earn goes to whoever the answer names.
of Google searches ended without a click in early 2026, up from roughly sixty percent two years earlier.
Similarwebdrop in organic click rate on queries that showed an AI answer, from 1.76 percent down to 0.61 percent.
Seer Interactive, 3,119 queriesmore organic clicks for brands that were named inside the AI answer than for those that were not.
Seer Interactiveof traditional search traffic is forecast to move to AI tools by the end of 2026.
Gartner forecastMost of what gets marketed as AI visibility work is one of these two. We ship the first one for a reason that has nothing to do with AI, and we leave the second alone.
Ahrefs took 1,885 pages that added structured data and compared them against 4,000 matched pages that did not, measuring AI answers before and after. No meaningful lift. Pages cited by AI are more likely to have schema, which is where the sales pitch comes from, but that is a correlation and it points the other way: good pages tend to have both. We still ship schema on every build, because it earns rich results in normal search. We will not price it as an AI feature.
The idea was a file at the root of your site telling models how to read you. Adoption reached about ten percent, and roughly ninety seven percent of those files were never fetched by anything. Google has said plainly that it does not use them. If a proposal lists this as a deliverable, you are looking at a line item that costs you money and changes nothing.
The list runs in the order the evidence supports. The build comes first, because everything under it is wasted if a crawler cannot read the page.
Pages arrive as finished text, not as an empty shell waiting on a script. You get this the same way you get pages that work on a phone, because it is how we build, and it never shows up as a line on your invoice.
Each page states the thing plainly in its opening lines, in sentences that survive being lifted out and quoted somewhere else.
Related pages link to each other by name instead of sitting alone. Pages inside a cluster get cited several times more often than pages with nothing pointing at them.
Schema on every build, because it earns rich results in ordinary search and that is worth having. You will never see us bill it as the thing that gets you into an AI answer.
Roughly half of the material these systems quote was published inside the last few months. Pages that sit untouched for two years stop getting used, so we revisit the ones that matter.
We run the real questions your buyers type, across the major assistants, on a repeating schedule, and keep the record. Attribution will not be clean. You will at least be looking at what came back instead of guessing.
Baseline first, because you cannot talk about movement without a record of where you started.
We put your buyers' real questions to the major assistants and record what comes back today, including who gets named instead of you.
We read every page the way a crawler takes it. This is a five minute check, and most of the time it is where the problem turns out to be.
Rebuild what arrives empty, rewrite the pages that never answer the question, and link the related ones together so none of them sit alone.
New material on a schedule, the old material revisited, and the same questions run again so you can see what moved and what did not.
No, and nobody can. There is no submission form, no placement to buy, no setting anyone controls. We can promise your pages will be readable, specific, current, and linked, which is the material every answer gets assembled from. Anyone selling you a guaranteed mention either misunderstands the system or is counting on you to.
They overlap, but they are not the same job. Classic search work is about ranking a link so somebody clicks it. This work is about being readable and quotable enough that the answer names you when nobody clicks anything. Google's own crawler renders JavaScript, which is exactly why a site can rank there and be invisible to the assistants. Most businesses want both. See the SEO page.
Run the ten second check before you spend anything. Put view-source: in front of your address and look for your own headline. If it is there, your build is fine and the work is content and structure. If it is not, no amount of writing fixes it from the outside, and the build is the conversation. See the websites page.
The readable part lands right away, since the next crawl gets a page it can use. Everything after that runs on how often these systems refresh and how often you are mentioned elsewhere, neither of which anybody schedules. We will not put a date on it, and you should be wary of anyone who does.
Imperfectly, and we will say so. No clean referral report exists for a conversation that happened inside a chat window. We run the same questions on a schedule, keep the record of who gets named, and watch direct and branded traffic next to it. Measurement here is rough, and anyone who tells you otherwise is selling you the comfortable version.
We will run your buyers' questions, show you what comes back, and read your source the way a crawler does. One call is enough to find out where you stand.
Readable first, then worth quoting