Conversational search is what happens when people stop typing keywords and start typing sentences. They ask a whole question. Then they ask another one about the answer they just got. The search box has turned into a conversation, and a page written for two word queries sits outside it.
None of this is a new channel to buy. It is a change in the shape of the question, so it changes which page wins. This article covers what really changes, what to do about it, and which parts of your current work already hold up.
What conversational search actually is
Conversational search covers any search where the question sounds like speech and where the person expects to keep going. A chat assistant is the obvious case. Google belongs here too, since a long question can return a written summary above the links.
Two features define it. First, the query is long and natural, so it carries context that a keyword never carried. Second, the session carries context, so the next question leans on the last one.
Conversational search is one slice of a wider change in how machines answer questions. If you want the whole picture first, read generative engine optimization and come back here for the question side of it.
How a keyword turns into a sentence
An old query looked like "boiler repair london". A conversational one runs closer to "my boiler is leaking and there is no hot water, who can come out today".
Look at what the second version carries. It names the problem, the urgency and the situation. None of that sat in the keyword, so the searcher used to supply it by scanning ten results.
Now the machine does more of that scanning. It reads the whole sentence, works out what the person needs, and picks a passage that fits. So your page competes against a far more specific question than before.
| Keyword query | Conversational query |
|---|---|
| Two or three words. | A sentence, usually with a complication in it. |
| The reader supplies the context. | The question carries the context, so the match has to be closer. |
| Ten links compete for a click. | A short answer has room for few sources, so the rest go unread. |
| Short enough that a tool reports a figure. | Split across many phrasings, so a figure is often missing. |
| One query, one page. | One question, then a follow up. |
The second question is the one that matters
A conversation, by definition, does not stop at one question. Somebody asks, reads the answer, then asks again with something narrowed down. The opening question is usually research. By the time somebody is asking who does this nearby, the decision is closer.
Conversational search happens in turns
A normal run goes like this. Someone asks what a service costs. Then they ask what changes the price. Finally they ask who does it near them.
Each turn narrows the field. If your page only answers the opening question, you show up early and disappear later. Whoever covered the whole run stays in the conversation.
Answer the next question on the same page
So the practical move is simple. Once you have answered the question a page exists for, deal with the obvious follow up in the next section. That is one page doing the work of three.
Ask your sales team what people say next. Their list beats any tool, because it comes from real conversations rather than from search volume.
A worked example, with round numbers chosen to make the arithmetic easy rather than to describe a typical result. Say ten people ask an assistant about your service in a week. The assistant names your page in four of those answers. Two of those four visit the site, and one of the two enquires. To reach a second enquiry you either appear in more answers or give the visitors a better reason to act. Both routes are work, and they are different work.
Keyword research for conversational search
Keyword tools grew up around short queries. A tool reports volume phrase by phrase, and people word one question a dozen ways. So each version carries only a slice of the demand, and many show no figure at all. That does not make them worthless.
It makes the unit of work different. Instead of chasing one phrase, cover one question properly and then cover what a reader asks right after it.
Where the real questions come from
- Sales calls and enquiry forms, where people describe the problem in their own words.
- Your inbox, especially the questions you answer every single week.
- Live chat logs, since they capture phrasing nobody would type into a tool.
- Reviews, yours and your competitors', because people explain what worried them.
- The assistants themselves, when you ask them your customers' questions.
Group what you find by the job the person is doing, not by the wording. Ten phrasings of one worry belong on a single page. Two different worries need two pages, even when the words overlap.
What to change on the page
Put the answer in the first two sentences
A machine handling a conversational search needs a passage it can lift. Give it one. State the answer plainly, then explain it underneath for the reader who wants detail.
Nobody loses out here. A human skimming on a phone wants exactly the same thing.
Write headings as the questions people ask
A heading such as "Our process" tells a machine nothing. "How long does a boiler replacement take" matches a question directly. Headings are the cheapest structural change available, and they serve both readers at once.
Keep the facts checkable
Say who wrote the page. Give a date. Name the areas you cover and the things you do not do. These details look dull. Still, they are what separates a claim somebody can check from one they have to take on trust. There is more on that side of the job in writing content for AI search.
Where conversational search changes nothing
Plenty of advice suggests you rebuild the site. You do not. The groundwork is the same as it always was.
Your pages still have to be crawlable. They still need to load quickly. A product page still needs a price and a way to buy. If a machine cannot fetch the page, nothing else on this list matters.
Two more things are worth saying plainly. A longer page is not a clearer one, and a chattier tone is not a clearer one either. In short, conversational search asks for a plain answer rather than more words.
Measuring conversational search without query data
Here is the uncomfortable part. Assistants hand you nothing like a keyword report today. You cannot see the sentence somebody typed, and usually you cannot see the follow up either.
So measurement moves to two questions. Do you appear when somebody asks? Then, what happens to the people who do arrive?
Testing by hand is crude, yet it is honest. Ask an assistant the ten questions your customers ask most. Write down who it names. Repeat that every month and watch the direction. Our piece on measuring LLM visibility sets out a fuller version of the same routine.
Watch your own analytics for a second signal. Not every visit carries a label saying where it came from. So treat what you see as a floor rather than a total. When an assistant names one exact page, the visit lands there rather than on your homepage. A conversational search then shows up as a deep entry rather than a homepage one.
Where voice fits into this
Spoken questions were conversational long before chat assistants arrived. Somebody asking a speaker for the nearest open pharmacy phrases it as a sentence, not as a keyword.
The overlap is large, so most of this article applies. Still, the reasons to bother differ by business, and we work through those in our piece on voice search.
The habits that waste the effort
- Stuffing questions into headings and then never answering them underneath.
- Rolling every question into one enormous page, which is different from covering one question and its follow up.
- Burying the answer beneath three paragraphs of background.
- Chasing volume figures for phrasings no tool will ever report.
- Ignoring the follow up, because that is usually where the buying question sits.
- Rewriting good pages in a chatty voice, which changes the tone and not the substance.
Where to start if you have one afternoon
Pick your three most valuable pages. Write down the question each one exists to answer. Then check whether the page answers it in the opening lines, and fix the ones that do not.
After that, add the first follow up question to each page as a real section. Finally, ask an assistant those same questions and note who it names today. That gives you a starting point to measure against next month.
None of this needs a budget. It needs a decision about which questions you intend to own.
Frequently asked questions
What is conversational search in simple terms?
Conversational search is search where somebody writes or speaks a full question and then keeps going with follow ups. Instead of typing two words and scanning links, they describe the situation and expect an answer back. The practical effect is that pages compete passage by passage. A page that answers one clear question in its opening lines has a far better chance than a page that circles the same subject for six paragraphs.
Is conversational search the same as voice search?
Not quite, although the two overlap heavily. Voice search is conversational by nature, because nobody speaks in keywords. Conversational search also covers typed questions in a chat assistant, and long typed questions in an ordinary search box. The work is mostly shared: answer plainly, structure the page around real questions, and keep your business details consistent. What differs is the setting, since somebody speaking to a device is often away from a screen.
Do keywords still matter for search?
Yes, they still matter. A keyword tells you which topic people care about and roughly how many of them care. What changes is the unit you work in. A keyword now sits inside a question, and that question sits inside a run of follow ups. So use keywords to choose the subject, then use real customer questions to decide what each page has to answer.
How long should an answer on the page be?
Give the direct answer in a sentence or two, then explain it properly underneath. Length on its own wins nothing. A machine looking for a passage to quote wants a short statement that stands alone, while a reader who has already decided wants the detail below it. If you cannot state the answer in two sentences, the page is probably trying to answer two questions at once.
Is an FAQ block worth adding for follow up questions?
Only where the questions are real ones. An FAQ block is a tidy way to cover follow ups that do not deserve a section of their own. It becomes padding when the questions exist to hold keywords rather than to help somebody. A good test is whether a customer has ever asked it out loud. If nobody ever has, leave it out and use the space for something else.
An assistant is giving people the wrong details about us. How is that fixed?
Fix the sources rather than the answer, because you cannot edit what an assistant says. Check the pages it is most likely to read: your service pages, your listings, and anywhere else your details appear. Make sure the name, the service area and the things you do not do all say the same thing. Then re-run the question in a few weeks, since an answer drawn from old pages keeps matching them until they change.
Can I track conversational search in my analytics?
Partly. Analytics can show visits that arrive from an assistant, where the visit carries a label saying where it came from. What it cannot show is the question somebody asked, or the answer they read. So pair that report with a manual test: ask the assistants your customers' questions and record who they name. Conversational search needs both halves, since the report measures arrivals and the test measures visibility.
Does conversational search help a small local business?
It can, because the person answering the phone usually knows the real questions. A specific answer is also easier to write when you do the work yourself every day. Keep your name, address and service area consistent everywhere. Answer the awkward questions competitors avoid, and say plainly what you do not do. Conversational search suits that kind of plainness, since a clear answer is the easiest thing for anyone to repeat.