Research demand outside Google

Search behavior spread across more surfaces than keyword research did. The standard workflow still assumes a query typed into an engine, a ranking, and a click you can count afterwards. A lot of discovery now finishes without all three: inside Pinterest's image grid, inside LinkedIn's people search, inside YouTube, and inside assistants that resolve the question and hand over no visitor at all.

A keyword tool shows none of it. That does not make the demand imaginary, and it does not make the research impossible. It means the method has to change in specific places, and this guide sets out where.

The part of the method that travels

The questions are identical everywhere. What does this person want, how do they phrase it, and how close are they to acting? Customer language answers all three whether you lifted it from a sales call or from a comment thread, and conversation seeding works on every surface for exactly that reason.

What does not travel is the measurement. No alternative platform hands over a query report, and none publishes a volume figure worth planning a quarter against. Remove those and the two things most workflows rest on stop working: sorting a list by a number, and confirming afterwards that a change did anything.

So you substitute. The platform's own search suggestions replace the volume column. What currently ranks there replaces the difficulty score. And a branded-search line replaces attribution. All three together are enough to work with.

Pinterest rewards the pin, not the follower count

People use Pinterest's search box the way they use an engine's, and the platform returns pins ranked against what they typed. Whoever owns the pin is usually invisible to them, which is why a small account can rank: the audience is not the input the ranking cares about.

That changes what you optimize. A pin has a title and a description. The image can carry text. The board has a name and a description, and the profile has an about field. A query has somewhere to sit in each of those, and the platform generally needs it in more than one before it is confident what the pin shows and what the account is for.

It is page optimization with different slots and the same rule underneath: a surface has to understand both the thing and its context before it will serve either to a searcher.

On LinkedIn, search reaches the profile first

What people do most on LinkedIn is open somebody's profile, which makes the profile the ranked asset and the feed a way of distributing it. That is roughly the reverse of where most members spend their effort.

Two fields do the matching. The headline and the profile's about text are what LinkedIn has to work with when somebody searches, and they are also the first thing a visitor reads. Both should be written in the vocabulary an outsider reaches for when naming that kind of problem. A headline stating an internal job title matches nothing anyone types.

Treat it as a positioning surface rather than a publishing one: a short list of the phrases your buyers use, a look at which profiles already come back for them, and a rewrite against the same vocabulary. It is a small amount of work compared to content, and it fails for the reason pages fail: writing for people who already know you.

Assistants finish some queries without a click

The third change is not another place to rank. It is a broken link between two things SEO treats as one. When an assistant can settle a question outright, a page can hold the top position and still receive nothing from it.

Commercial results had already moved this direction, because paid blocks and answer modules push the first organic listing well down the screen. First in the ranking and first on the screen stopped being the same thing, and only the second one pays.

You cannot undo it, but you can price it in. Questions an assistant can close in two sentences are worth less than their volume implies. Queries that need a price, a login, a tool, a location or a person still finish with a click, because an answer alone does not complete them. Splitting your own list that way shows how much of it is exposed before you build on any of it, and it changes the numbers you feed into sizing.

Research without a volume column

Working qualitatively still gets you a usable list.

1. Type the seed into the platform's search box and record every suggestion it offers. That is the platform reporting what its users look for.
2. Open what ranks for those suggestions and read the titles and opening lines. The phrasing there is what has been rewarded so far.
3. Look for first-party reporting before trusting either of the above, since a few surfaces do hand over something.
4. Track branded search. When a platform keeps the click, interest still leaks out later as somebody typing your name.

The fourth is the only feedback you get, and it is weak on its own and useful as a trend. Commit to a surface that sends nothing back, and a branded line moving inside your own Search Console over the same period is the nearest thing to confirmation available.

Do it with an agent

A demand baseline still has to come from somewhere. Google is an imperfect stand-in for what people want, and an odder one when the question gets asked somewhere else, but it is the only baseline anyone can buy. Connect the Doverity MCP and run this prompt.

Use the Doverity MCP. Run research_keywords on "[my topic]" and call
get_keyword_metrics on the output so every keyword has volume and
intent attached.

A. Sort the list by whether an assistant can settle it. Put queries
needing a price, a login, a tool, a location or a human at the top
and queries answerable in two sentences at the bottom, then tell me
what share of total volume lands in each half.

B. Use get_search_console_performance to give me my branded queries
and how they moved month by month over the past year. I am about to
put time into [platform], which keeps its click data private, so this
line needs recording first.

C. Rewrite my ten best non-branded keywords the way somebody would
phrase the same need inside Pinterest or LinkedIn: fewer words, more
visual nouns, more job titles.

Read part A first. It is the difference between a list that looks large and a list that still produces visits.

Keyword research beyond Google FAQ

Which surfaces are worth researching besides Google?

Whichever ones your buyers already use to look things up, which is a shorter list than the number of platforms that exist. Pinterest and LinkedIn carry clear intent and reward keyword work directly. YouTube, Reddit, and community forums carry intent that shows up later as branded search. Start with the one or two places your customers name when you ask them where they looked, rather than trying to cover all of them.

How do keywords work on LinkedIn?

Through the profile. Searching for people is one of the most common actions on the platform, so the headline and the profile's about text are what get matched against a query, and they should carry the vocabulary an outsider reaches for when naming that kind of problem. Posts then distribute a profile that already reads clearly; they do not substitute for one that does not.

Should informational keywords be dropped because of AI?

The ones that were only ever definitions have weakened, and the rest have not. Ask of each query whether an assistant can close it in two sentences. If it cannot — because it needs a price, a login, a tool, a location or a person — the click survives, and those queries are often the more valuable half of a list that volume alone would have sorted differently.

What replaces volume data on other platforms?

Three substitutes, none of them as good as the real thing. The platform's search suggestions show what its users look for. What currently ranks shows what the surface rewards. And a branded-search trend shows whether any of it is working. Used together they are enough to plan with.