Where your buyers discover new experts now

Four routes now bring a stranger to your name.

Someone asks a chatbot for a shortlist. Someone forwards your writing inside a private thread. Someone hears you on a show or in a newsletter they already read. Or a colleague says your name, and the person goes and checks.

The last route has worked roughly the same way for decades. Three of the four leave almost nothing behind in your analytics.

So the question of how buyers discover new experts has two answers. There is what the studies report, which is worth reading with care. And there is what you can see from inside your own business. That second view is narrower, and it has the advantage of describing your buyers instead of someone else's sample.

Chatbots now build the shortlist before anyone contacts you

G2 surveyed 1,076 business software buyers in March 2026. Half said they now begin research with a chatbot more often than with Google. That share was 29% seven months earlier.

The number that matters more for this question sits further down the same report.

Of those buyers, 33% bought from a vendor they had not heard of before the chatbot named it.

That figure is discovery in the plain sense. For that third, the chatbot did not speed up a search they were already running. It introduced a supplier they would not have found.

Read the size of the number with care. These are people buying software. Software comes with review sites, comparison grids, and structured listings that models read easily. Hiring a tax advisor or a litigation consultant runs through fewer of those. The direction transfers to expert work. The percentage does not, and quoting it about professional services stretches it past what it measured.

What follows for an expert is smaller than the headline and steadier. Being named inside the answer now does the work that ranking underneath it used to do.

The trackers disagree, and their products explain some of it

Yext analyzed 6.8 million citations from ChatGPT, Gemini, and Perplexity across two months of 2025. It found 86% of those citations came from sources a business already controls, mostly first-party websites and listings. Forums came in at 2%. Yext sells listings management.

Other trackers publish close to the opposite picture. Semrush's study of the domains chatbots cite puts Reddit at or near the top across several engines, and rival monitoring companies report the same ranking. Those companies sell visibility monitoring for that kind of citation.

Both sets of numbers can be correct. Yext ran location-specific consumer questions across four industries. The Reddit-heavy studies ran broad prompts across every category they could reach. Different questions produce different sources, and both results get published as the state of the web.

Take the lesson rather than either ranking. A channel study describes its own sample. That is why I check what practitioners report before repeating any tactic scored in public.

Most discovery now arrives with no return address

Chatbots strip the referrer on most of what they send. Private shares carry no tag at all: a link pasted into Slack, an email forward, a screenshot in a group thread. A podcast mention produces someone typing your name into a browser a week later.

All of that lands in your analytics as direct traffic, or as nothing.

The click numbers make the same point from the other side. Referral visits from chatbots are still a fraction of 1% of worldwide site traffic, tracked month over month. Buyers report being influenced by chatbots at rates many times higher. Both readings hold, because the influence and the click have come apart.

Your dashboard reports the channels that leave receipts. It stays quiet about the ones that leave impressions. A business funding only what the dashboard can see keeps paying for the visible half and quietly defunds the rest.

For expert firms, most introductions still travel through people

Hinge studied 495 professional services firms for its 2026 report and asked which techniques delivered the greatest impact. Hosted events led the list, tied with networking at conferences, with thought leadership on social platforms beside them.

Read that as what firms report about themselves. Buyers were not the ones asked, and an event is visible to a seller by construction.

The shape still fits everything above. Someone the buyer already trusts hands over a name, and the handoff happens in a room, a feed, an inbox, or an episode.

I produced the Deliberate Leaders Podcast for more than 200 episodes and read the pitches that came in. A dozen would arrive each week, and a few of those were good. The good ones had studied the show. They arrived with questions the host's audience would want answered. Discovery on a borrowed audience is granted by a person, and that person is protecting their own listeners first.

The newsletter deserves its own note, because it looks owned and its growth is not. On Substack, much of the early subscriber flow arrives through the platform's recommendation engine and its social feed. On beehiiv, the equivalent growth runs through paid and reciprocal recommendations between publications. The list you keep is yours permanently. The stream of new readers arrives on somebody else's terms, and those terms change without notice.

Being verifiable everywhere beats being dominant anywhere

You cannot rank the four routes for your own business with any confidence. The studies that promise a ranking are reading their own samples. What you can do is make every surface resolve to the same competent picture of you.

That means the site and the profile a buyer opens next. It means the listing a chatbot reads, the transcript of the talk, and the article a peer forwards. Someone who hears your name goes and checks. What they find either confirms the recommendation or ends it without a word to you.

Consistency is doing more work now than volume. A model assembling an answer about you rewards a record that agrees with itself. So does a person deciding whether you match the description they were given.

The one instrument you own is the question

Ask every person who reaches you how they found you, and ask it in two parts.

The first part is a short list to choose from: search, a chatbot, a person, a newsletter, a podcast, an event, social. The second part is a text box with a plain prompt. What did you ask, or where did you see me?

The second part carries the value. "ChatGPT" tells you almost nothing. "I asked which firms handle cross-border estates for American expats" tells you what to publish next. It also tells you which words your buyers use when they have no name to type.

Read the answers monthly and keep them somewhere searchable. A quarter's worth of free-text answers can be grouped and counted in a few minutes with any current chatbot. There is no longer a good reason to let them pile up unread.

What this reading will not tell you

Self-reported answers come with known faults. People name the last thing they remember, which flatters whichever channel arrived most recently. Some write "Google" for a route that ran through three places first. A firm taking a handful of clients a year gathers small numbers, and small numbers swing on one answer.

The platform figures also age faster than the buying habit does. Chatbot research rates moved by more than 20 points inside a year. The citation studies contradicted each other inside a single quarter. Treat any figure about assistant behavior as a reading from its date, and check it again before building a plan on it.

None of this settles which channel you should work. That decision follows the format you can sustain and the way your buyers verify a name. It answers differently for a solo advisor and a firm of thirty.

You will not find a ranked list of channels that is true for your business. The places that introduce you leave no record, and the studies that rank them are describing somebody else.

What you can have is smaller and firmer. The last twenty people who reached you can tell you how they got there. And the next person who checks your name will find a record that either holds up or does not.