How customers find you
in the age of AI.

01 / The introduction

A business can help answer the question and still miss the introduction.

The search is only three words: “commercial mortgage toronto.” In one Google screenshot, EquityRich occupies the first visible organic result. There is the company’s name, a blue link and a few lines describing its offer. Anyone interested can step through that small doorway into the company’s own account of itself.

This is the familiar arrangement behind search engine optimisation, or SEO: make a useful page discoverable for a relevant search, then give the person who arrives a reason to stay. The results page—often shortened to SERP—helps the searcher choose where to go. The business gets a chance to make its case.

From the archive

The familiar introduction

A second capture, for the same words, shows a more complicated introduction. Google has written an answer of its own. EquityRich appears in a supporting source card, but the company named in the visible answer is CMLS Capital. A reader could take in the explanation, remember the name in the paragraph and never investigate the source beside it.

The captures were made on different dates, and the second even shows Kansas as its location context despite the Toronto query. They are observations, not a controlled comparison. Their value is more basic: they make a distinction visible. Being used as a source and being introduced as a business are separate events.

From the archive

The source and the named business

For an owner paying to be found, that separation deserves attention. A company can contribute useful knowledge without receiving the visit it hoped to earn. It can also be named in an answer and gain an introduction that a website report never records. Neither possibility tells us what happened to sales. Both complicate the old habit of treating traffic as a complete account of discovery.

This is the change worth understanding beneath the acronyms and the confident predictions. Search is acquiring more of the work that happens between having a question and choosing whom to trust. The customer may still reach your website. By then, some of the conversation may already be over.

02 / The conversation

The answer becomes a place to spend time.

Consider a homeowner planning a kitchen renovation. This is an imagined example, but the questions are ordinary enough: what will it cost, which expenses tend to surprise people, and can the family live in the house while the work is done? Those questions precede the search for a contractor. They help the homeowner decide what kind of contractor to look for.

A set of links leaves much of that investigation to the reader. They open pages, compare claims and carry what they learn from one website to the next. A generated answer can gather an explanation in one place. A follow-up question can narrow it: suppose we keep the existing layout; suppose the house is a century old. Research starts to resemble a conversation.

The appeal is easy to understand. The reader gets to ask the question in the shape it has in their life. The trade-off is quieter. Someone—or, here, a system—has chosen which details to include, whose explanation to draw on and which names deserve space. Convenience gives those choices influence.

Google’s AI Overview puts a generated response inside the search experience. Look at the landscape-design capture below: Angkorscape appears in the answer and in a source panel. Unlike the mortgage example, the business receives a visible mention as well as a citation. The useful question is what role a company plays in the response, not simply whether its website appears somewhere on the screen.

From the archive

An answer with room for a business name

AI Mode makes room for a longer exchange. The capture below includes an answer, a map, sources and a field for the next question. Google explains that its AI search features may run related searches while assembling a response. A customer’s single question can therefore lead the system through more than one underlying search. Google describes how these features work.

From the archive

A place for the next question

ChatGPT search offers another setting for this kind of investigation. In our fencing example, it has organised businesses into a shortlist and supplied links. The list explicitly says it is alphabetical: the last company has not “ranked seventh” in the familiar SEO sense. Reading the answer closely changes what can honestly be claimed about it.

From the archive

A shortlist is not a league table

The vocabulary has followed the interfaces. AEO, or answer engine optimisation, describes work intended to help information appear in answers. GEO, or generative engine optimisation, focuses on visibility in generated responses. In practice, the terms overlap. They describe ambitions more neatly than they describe separate systems you can switch on.

SEO remains the foundation beneath much of this work: helping search engines access and understand pages, and making those pages useful for the searches they serve. Google’s introductory guidance sets out that basic job. The newer terms are useful when they sharpen a question—are we being cited, named or recommended? They become less useful when they make a familiar service sound newly mysterious.

For our hypothetical renovator, the practical problem is wonderfully concrete. Does the business have anything worth drawing on when a homeowner asks about an old house, a tight budget or the mess of living through construction? A slogan about quality leaves little to work with. A detailed account of an actual project gives both the homeowner and a search system something more substantial to examine.

03 / The missing visit

What becomes of the click?

There is a moment in any search when the person has enough. Enough to identify a material, understand a phrase or decide that a project is more expensive than expected. If an answer supplies that stopping point, opening a website becomes optional.

The concern for businesses is that the visit they used to earn was sometimes a by-product of delivering the explanation. The reader came for an answer and, along the way, encountered the company. An answer delivered on the results page can satisfy the first need while removing the occasion for that encounter.

Pew Research Center studied the browsing activity of 900 U.S. adults in March 2025. In searches classified as displaying an AI summary, users clicked a traditional result on 8% of visits, compared with 15% when no summary appeared. Clicks on links inside the summary occurred on just 1% of visits with a summary. The study was published in July 2025.

Those figures give the concern weight, but they need their setting. This was observed behaviour, not an experiment assigning identical searches to two different screens. Pew collected the result pages in April, after the browsing period. The searches with summaries could also differ from those without them. The comparison describes a pattern; it does not isolate the summary as the sole cause.

Ahrefs approached the question from another direction. Its February 2026 analysis compared desktop data for 300,000 keywords across December 2023 and December 2025. It used the decline among keywords without AI Overviews to estimate what click-through might otherwise have been for those with them. For the top-ranked page, the observed rate was 58% below that modelled baseline.

The important word is “modelled.” This is an estimate built from a comparison, not a meter measuring how much traffic AI removed from your business. Its practical implication is nevertheless serious: a high organic position may earn a different share of attention when the page around it changes.

There is no single business experience concealed inside these averages. Someone looking up a definition may finish immediately. Someone choosing a contractor for a large project may keep investigating long after reading a summary. Both are searches; the economic meaning of their clicks is different.

That is why a blanket promise about replacing lost traffic with AI visibility deserves the same scrutiny as a blanket prediction of collapse. The useful work begins with the particular questions your customers ask—and with knowing which visits were valuable in the first place.

04 / The misleading average

A report can look worse while showing you more.

Search reports have a peculiar talent for making a complicated situation look settled. One number rises, another falls, and a colour supplies the verdict. Green feels like progress. Red asks for an explanation.

Take the historical report below, from the CloudCure case study. Clicks rose from roughly 1,310 to 6,420. Impressions climbed from about 4,500 to 208,000. Yet click-through rate fell from 29.2% to 3.1%, and average position moved from 8.2 to 27. A reader looking only at the last two measures could mistake a much larger search presence for a simple retreat.

From the archive

More visits, a lower click-through rate

An impression records an appearance of a link under Google’s counting rules; it is not proof that a person noticed or remembered the business. A click records someone following a link. Click-through rate, usually shortened to CTR, divides clicks by impressions. Average position summarises where results appeared, with rules that vary across search features. Google explains the definitions and counting rules.

The arithmetic in this report is revealing. The smaller period produced about 29 clicks for every hundred impressions. The larger one produced about three. But those three came from a vastly larger pool, so the total number of visits grew. The rate fell as the volume rose.

That does not explain why the pool changed. One possibility is that the site began appearing for many more searches, including ones where it occupied lower positions. Establishing that would require the underlying query and page data. The screenshot alone gives us the totals, not the history that produced them. These historical totals tell us nothing about an AI-search effect; this is a lesson in reading a report.

Now consider a different set of numbers. A site earns 400 clicks from 10,000 impressions. Later, it earns 300 from 20,000. Twice as many impressions, fewer visits. The result can be perfectly real without telling us whether the business is becoming easier for the right people to find.

If the additional impressions concern services the company cannot supply, their commercial value may be slight. If the lost clicks came from a valuable service page that slipped behind competitors, the loss deserves attention. If positions stayed similar but an answer began satisfying the question above the links, the explanation may lie in the results page itself. An aggregate graph cannot choose among those stories.

The investigation is to compare like with like: the same important services, locations, pages and groups of queries over sensible periods. Separate searches for your name from searches by people who have not chosen a provider. Look at the screen those customers actually encounter. Then bring in whatever you know about the enquiries and customers the visits produced.

This is where impressions become more useful—and where it is easiest to ask too much of them. They can expose growing reach that clicks alone miss. They can help locate a change in attention. They cannot tell you that a mention persuaded someone, that a citation was read or that an enquiry will arrive later.

Google includes traffic from its AI search features within Search Console’s Web reporting. An increase in that report therefore does not, by itself, identify which impressions came from AI answers. Google documents this reporting scope. A screenshot of a citation is a separate observation, not a missing column you can simply add to the graph.

For a business, visibility, response and revenue belong in the same conversation. They answer different questions. Visibility concerns the chance to be encountered. Response shows someone doing something with that opportunity. Revenue tells you whether the work eventually supported the business, insofar as you can trace it. Keeping the distinctions intact makes the report more useful than declaring any one of them the new king.

05 / The useful detail

What can a customer learn that only you could tell them?

Return to the renovation company. Almost any competitor can publish an article listing the advantages of a new kitchen. The owner has access to a different kind of knowledge: what happened when a wall came down on a particular job; why a proposed layout changed; which decision saved money and which apparent saving proved expensive.

A project account built from those facts does several jobs at once. It explains a problem a homeowner might recognise. It gives the reader evidence of how the company thinks. It supplies detail that can be checked against photographs, specifications or the people involved. Even a reader who disagrees with the decision has learned something useful about the business.

The principle travels, although the evidence changes. Imagine an electrician explaining an EV-charger installation. “We install chargers” establishes the service. An account of the site assessment, equipment and limits of the job helps a customer understand what an installation involves. The company should explain the work it actually performed; it need not pretend every house will present the same conditions.

For an accountant, the useful detail might be how an engagement works: which records the client needs to supply, where responsibility changes hands and what the quoted fee covers. For a specialist retailer, it might be a comparison of two products using published specifications and firsthand observations, including the circumstances in which the cheaper one is sufficient.

These are examples of what to document, not claims that any particular page will secure an AI mention. Their value begins with the customer. They replace the demand to trust a company’s adjectives with material a person can inspect.

This also explains why producing more pages can feel busy while adding very little. A tenth account of the same general advice does not automatically answer the question the first nine avoided. The more useful editorial question is: what does a person need to understand before making this decision, and what evidence do we have that would help them understand it?

For a smaller business, that question has an encouraging answer. You may not possess a national brand’s resources. You do possess the particulars of your own work. Treating those particulars seriously is a more credible starting point than trying to sound authoritative about everything.

06 / The work underneath

The new language still needs something solid beneath it.

Useful evidence must also be accessible. A careful project account hidden behind a broken page, an unexplained service or an inaccessible image cannot do its full job. This is where the technical and editorial sides of SEO meet: the business needs to make its information available and give that information a coherent meaning.

Google says a page must be indexed and eligible to appear with a snippet to be a supporting link in its AI features. It does not require special AI markup or a new machine-readable file for that purpose, and eligibility does not promise selection. Its guidance is explicit about those requirements.

That makes a sensible first examination fairly concrete. Can the important pages be reached and understood? Do they accurately describe what the business offers, where it operates and what supports its claims? Are useful explanations attached to the relevant services, or stranded in an unrelated corner of the site? The answers identify actual work rather than a package of new terminology.

The same care should apply to machine-written material. Generative tools can help organise and express knowledge, but fluent prose is easy to produce without adding knowledge at all. Google’s guidance warns against generating pages at scale without value for users. Its advice concerns the usefulness of the result. For the owner commissioning the work, the immediate test is simpler: what did this page teach a prospective customer that our existing pages did not?

Observing AI answers should be equally disciplined. Choose a manageable set of real buying questions and record the platform, wording, date and relevant location or conversation context. Save what was shown. Distinguish a link used as a source from a business named in the answer, and both from an explicit recommendation. Repeat the observations rather than turning a favourable screenshot into a permanent claim.

ChatGPT search can use context, including approximate location, when producing results. OpenAI also describes how its search crawler relates to website eligibility. Its documentation is a useful starting point. Those details matter when two people ask apparently similar questions and see different answers.

The resulting record will be incomplete. It can still be useful: a description that is consistently wrong deserves investigation; a valuable service that is poorly explained deserves better material. A handful of searches cannot stand in for the experience of every possible customer.

07 / Back to the business

Ask for an explanation you can use.

The next SEO meeting need not begin with a prediction about the future of search. Begin with a part of the business you want to grow. Put the relevant pages, search results and enquiry evidence on the table. Ask what a prospective customer can learn, where the company is being encountered and what remains unknown.

There should be room for an honest mixed result. Perhaps more people can find a service, but the enquiries are poorly matched. Perhaps visits have fallen while qualified enquiries have held up. Perhaps an AI answer names the business, but nobody can yet connect that appearance to a customer. Each situation calls for a different next step. None improves when uncertainty is concealed behind a larger number.

Agree on one gap worth addressing and a date to examine what happened. A review date commits people to learning from the work; it cannot oblige a search engine to deliver a result. The questions below are meant to help make that conversation specific.

Think again of the mortgage company in the source card. Its presence tells us something real, but smaller than the whole story. We can see that its page was offered as a source. We cannot see whether the reader noticed, clicked, remembered or called.

The future of finding customers will contain more of these partial views. The task is to read them carefully, make the business worth investigating and keep asking what happened after it was seen. A useful explanation, a well-documented project, an honest account of a limitation: these give a prospective customer something to judge. They remain worth making, wherever the introduction happens.