AI search visibility

It is whether AI assistants name your business when someone asks them a buying question. Instead of ranking in a list of ten links, you are either quoted in the answer or you are invisible. It is measured by tracking real prompts across the engines and recording who gets cited.

Ask it, then ask it the questions your customers would actually ask, then do the same in Perplexity and Gemini. You can do this yourself in an afternoon and we recommend you do before hiring anyone. What we add is doing it systematically across dozens of prompts, repeatedly over time, so you can see change rather than a snapshot.

It overlaps, and anyone who tells you otherwise is selling. The technical foundations are shared. What is genuinely different is the unit and the goal: prompts instead of keywords, being quoted instead of being ranked, four engines instead of one, and the fact that a citation often produces no click at all. Sold as SEO, it gets measured with the wrong instruments.

Because assistants prefer sources that are not the company being asked about. Ask one who the best supplier in a sector is and the citations tend to land on comparison pages, directories, roundups and independent sites. A brand writing about itself is a weak source in that stack, which puts a real ceiling on what content on your own domain can achieve on its own.

No, and nobody can. The engines change their sourcing regularly, answers vary between users, and none of them publish a ranking algorithm to optimise against. What we can guarantee is that you will know exactly what they say today, what changed since last month, and what is being done about it. If someone guarantees you a placement, ask them how they intend to control a model they do not own.

Usually an answer box. Google folds AI Overview appearances into the standard Search Console performance report without a separate filter, so a page can hold its position, get shown more, and be clicked less, because the answer was delivered above it. Read as totals those impressions look like growth. Read as a ratio against clicks over time, the gap is the clearest signal you have that this is happening to you.

ChatGPT, Perplexity, Gemini and Google AI Overviews, with Claude added where the question suits it. Coverage differs by engine and we tell you where it is partial rather than presenting a tidy dashboard that implies more certainty than exists.

No, and content on your own site is often not the strongest move available. We also rebuild the pages you already have into a form the engines can quote, fix the technical and entity signals, and where it makes sense build an independent property that earns citations in its own right and routes readers to you. We run sites like that ourselves, including snowverdict.com.

We record what the engines say about a page before it exists, then re-check weeks later, which is the only way to know whether anything moved. The most recent round covered 25 published articles and pages against their own pre-publication baselines, across 309 separate question-and-engine checks. Across that round the client domain appeared as a source about twice as often as at baseline, and the share of those appearances the model actually used in its answer, rather than merely retrieving and ignoring, rose from roughly a third to roughly three quarters. On VanBaltic, a campervan rental company, we ran a controlled test on the Google side of the same work: pages the new articles linked to grew impressions by 111% over four weeks, while un-linked pages on the same site grew by 9%. Because both groups sit on one site in one season, seasonality cannot account for the difference. One caveat we would rather state than have you find: those checks are a single sample per question on a single day, and AI answers vary between runs, so we treat them as a demonstration of method rather than a guarantee of outcome.

Articles, overwhelmingly, and it surprised us too. In our most recent measurement 30 of 31 citations landed on articles that answer a question, while 14 commercial route and service pages produced exactly one citation between them. Every broad commercial query of the "best supplier in city X" kind returned nothing for those pages, before and after. The practical consequence is that trying to get your sales pages quoted is mostly wasted effort. The work that pays is earning the citation with a page that answers something, then routing the reader from there to the page that sells.

Less time than most people assume, and it varies enormously by engine. SISTRIX tracked 82,619 prompts across seventeen weeks and found cited sources being replaced at roughly 5% a week in Google AI Overviews, 56% in Google AI Mode and 74% in ChatGPT Search. The more useful finding sits underneath that spread: an answer usually holds a stable core of one to five domains that persists for months, with a rotating cast of others around it, and 53% of AI Overview questions changed no source at all over the whole period. So there are two very different outcomes. Reach the core and it holds. Land in the rotation and you are renting the position. Which of the two you are in cannot be seen in a single check, which is why we read a citation as a rate over several weeks rather than a yes on the day we happened to look.

Ongoing, and we would rather explain why than pretend otherwise. Two things move underneath you: the engines change how they choose and weight sources, and your competitors keep publishing and improving. On the more volatile engines the sources cited for a question can turn over week to week. That is not a reason to panic, and we will not sell it as one, because the work itself does not decay. It is the reason a position checked once a year is a position you no longer know. The honest shape is a build phase of roughly ninety days followed by maintenance and expansion, with the measurement deciding which of those a given month needs.

No, and this is the part most agencies are vague about. Two things are moving here and they are easy to confuse. The asset is durable: a page rebuilt so a model can quote it stays quotable, structured data stays valid, and a citation you have earned does not expire on a billing date. The position is volatile: whether you are being quoted this week is a separate variable that moves on its own. What you lose by stopping is the watching, so you would no longer know when an engine started citing a competitor instead of you, or began stating something about your business that is out of date. The asset stays. The visibility into it does not.

Often not. An AI citation that produces no click leaves almost no trace, and attribution here is genuinely weak. That is the strongest argument for tracking the answers directly rather than waiting for traffic that may never be attributable, and it is why this is priced on visibility and accuracy and never on promised revenue.

A piece is graded on two axes, not one: whether it got cited, and whether it earned traction. Cited with traction is a win. Either one on its own is a partial, and both are results rather than failures. A fail has a defined ladder - clickthrough pass, answer-block rebuild, re-target, retire - with two attempts and then a decision, so remediation is planned work rather than an apology and sunk cost has nowhere to hide.