Home / AI visibility
AI visibility in Andover
A growing share of buyers now ask an assistant instead of scanning a results page. The answer they get names some businesses and not others and that selection follows rules you can work with.
This page explains those rules mechanically. No guaranteed outcome sits at the end of it and anyone offering you one is guessing.
What these systems actually do
The main assistants do not answer from memory for anything current, local or specific. They run a search, retrieve a small number of pages, and compose an answer from what those pages say.
That single fact governs everything else on this page. A site that is never retrieved cannot be cited. A site that is retrieved but ambiguous gets skipped in favour of a source that is easier to quote.
| Surface | How it finds a business | What it rewards |
|---|---|---|
| Google AI Overviews | Draws on Google’s own index and typically summarises pages already ranking for the query, with links out to a few of them. | Ranking plus a clean, extractable passage |
| ChatGPT search | Runs a live web search for current or local questions, then writes from the pages it retrieves and cites some of them. | Being retrievable and unambiguous |
| Perplexity | Retrieval first by design, with visible citations against most statements in the answer. | Specific, checkable statements |
| Copilot | Built on Bing’s index, which means Bing coverage matters here in a way it does not elsewhere. | Bing indexation and consistent details |
how it works Each row describes how the named system retrieves and composes an answer, as explained in the copy above. None of it is a claim about your specific results.
What makes a passage quotable
A model quoting your page is doing something narrower than a person reading it. It needs a passage that stands on its own, states a fact plainly, and does not depend on the three paragraphs above it for meaning.
Marketing prose fails this test almost completely. A sentence like "we deliver bespoke solutions tailored to your needs" contains nothing that can be lifted into an answer. It asserts nothing.
This is why the writing on this site looks the way it does: short paragraphs, a plain answer directly under each heading, and specifics rather than adjectives. The format is the argument.
What can actually be changed
- Be retrievable at all
Crawlable pages, correct indexing directives, and a site that does not block the crawlers these systems use. A surprising number of sites quietly exclude some of them.
Bing indexation is checked separately. Copilot depends on it and many sites have never looked.
- State the facts unambiguously
Business name, what you do, where you do it, how to reach you, all stated plainly in text and mirrored in structured data rather than left to be inferred from a logo.
- Write extractable answers
Each heading phrased as the question a person would actually ask, with the opening sentences underneath forming a complete answer on their own.
- Clean up the contradictions
Old directory entries, abandoned profiles and outdated details are the usual reason an assistant says something wrong about a business. Fixing them is unglamorous and effective.
- Record what the answer says, with a date
The same questions asked on a fixed schedule, with the responses saved verbatim. Without that record, nobody can tell whether anything changed.
How much traffic do AI assistants actually send?
This is the AI referral record for one site in my own portfolio, over the 90 days to 19 July 2026. It is a small channel next to Google and it did not exist as a channel at all two years ago.
AI assistant referral sessions, 90 days to 19 July 2026
Portfolio site, health and wellness sectorsourced figureGA4, 90 days to 19 July 2026, pulled 20 July 2026.
A site in my own portfolio, not a client engagement, shown by sector rather than by name. Results vary with niche, competition and starting position.
Perplexity here is the sum of two source labels that GA4 reports separately. One SEO tool that GA4 groups with these sources is deliberately excluded. Counting a crawler as an assistant would overstate the figure.
Where this bites locally
Ask an assistant for a supplier of a specific industrial service near Andover and the answers are frequently thin, hedged, or drawn from a directory rather than from any supplier's own site. That is a gap and it is currently cheap to fill.
The pattern is strongest in exactly the sectors the town runs on: distribution, light industrial services, agricultural supply and professional practices. These are categories where buyers ask a question before they know which company they want.
It is weakest for businesses whose customers already know the name they are searching for.
Asked most often about AI visibility
How do AI assistants decide which businesses to name?
They run a search for anything current or local, read a handful of the returned pages, and write an answer from what those pages say. Being named therefore depends on being in the results they retrieve and being written in a way that is easy to quote accurately.
Is this just SEO with a new label on it?
It overlaps heavily but it is not identical. Ranking rewards a page that satisfies a searcher who clicks through. Citation rewards a passage that a model can lift without ambiguity, putting far more weight on unambiguous facts and self-contained answers.
Why do assistants sometimes state things about a business that are out of date?
They are reconciling several sources of different ages. Where those sources disagree the model often picks the one that appears most often rather than the one that is most recent. Old directory entries and abandoned profiles do real damage here in a way they never did in ordinary search.
Is there enough traffic in this to be worth the effort?
On its own, for most local businesses, not yet. It is worth doing because the same work improves ordinary search at the same time and because the answer an assistant gives about your category is being formed now whether you participate or not.
Can you guarantee that ChatGPT will mention my business?
No. These systems are not deterministic and the same question can produce different answers on different days. What can be done is to remove the reasons you are currently unquotable and to record, with dates, what changes.
Find out what the assistants currently say
The Sprint asks the questions your buyers ask, records the answers verbatim with dates, ships five fixes, and sets a re-check date.
No guarantee of citation is offered, because none can honestly be given.