Keywords are questions. Hashtags are rooms.
Describe the account and pick the platform. You get the five phrases that describe it, eight long-tail phrases a specific person would type, six questions typed into the search box, ten hashtags and three topics people ask about that few accounts answer - written for the app's own search, not for Google.
The Social Keyword Finder suggests keywords and phrases to write around for the audience you describe, and lets you download the list to sort and plan from. The suggestions follow the language of the topic, and each one should be checked against real search volume before you rely on it.
Inside the app, search is a different machine
Google answers a question with a page. Instagram answers it with a video, and it decides which video by reading the caption, the on-screen text, the spoken words and the topic the account has taught it. That means the words a creator uses are not decoration; they are the only index the platform has.
The phrases people type there are also different. Nobody searches \"meal prep ideas\" on TikTok; they search \"lunch for work that does not need a microwave\". The long, slightly awkward phrase is the real query, and it is the one a keyword tool built for websites will never show you, because the search volume is too small to register and the intent is too specific to matter elsewhere.
Hashtags are a different system again. They used to be a discovery mechanism and are now mostly a topic label: useful for telling the platform what a video is about, useless as a reach strategy. Ten relevant tags do that job; thirty tags copied from a list do not, and they make the caption look like spam to the humans reading it.
Three layers, and an account needs all of them
The wide words bring strangers, the narrow words bring the people who save the video, and the questions bring the people who were already looking for an answer. An account that only uses the wide layer grows slowly and generically; one that only uses the questions never reaches anybody who was not already searching.
The wide words bring strangers, the narrow words bring the people who save the video. An account needs all three layers, in that order of frequency.
The percentages are the shape of a healthy mix, not a rule: most posts should sit in the middle layer, because that is where a specific person recognises themselves in the first line. The wide layer is for reach, and it works once the account already has a topic; without one it produces views that do not convert into followers.
Where each phrase belongs
| Layer | Example | Where to put it | What it does |
|---|---|---|---|
| Core | sourdough | first line of the caption, bio, profile name | teaches the account's topic |
| Long-tail | sourdough starter smells like acetone | caption, first two seconds of on-screen text | matches a specific search |
| Question | why is my starter not rising | the hook, said out loud | matches how people type |
| Hashtag | #sourdoughstarter | end of the caption, three to ten of them | labels the topic |
| Gap | what to do with a starter you forgot for a month | a video nobody has made yet | earns saves, not views |
One phrase per video, used in three places, beats ten phrases used once. The platform reads the caption, the spoken words and the on-screen text, and a phrase that appears in all three is a much stronger signal than a list of tags at the bottom.
Turning the list into a calendar
Fix the core phrase
One phrase that describes the account, used everywhere: bio, profile name, the first line of every caption.
One long-tail per post
Eight phrases, eight posts. Each becomes a title, a hook and an on-screen line - not a hashtag.
Answer a question on camera
The six questions are video ideas. Say the question out loud in the first two seconds, then answer it.
Fill one gap a month
The three gaps are the videos that earn saves instead of views, and the ones the algorithm has not seen.
Three habits that beat a keyword list
Platforms transcribe speech. A phrase that is spoken in the first two seconds is indexed, and it is also the fastest way to make the video understandable with the sound off. Reading a list of tags silently does nothing.
An account that posts about food, travel and productivity teaches the platform nothing, so each video is shown to nobody in particular. A narrow topic means the second video is shown to the people who watched the first - which is the whole mechanism.
Type the core phrase into the app's search box and write down what it suggests. Those suggestions are the phrases people typed, ranked by frequency, and they are free. The tool gives you a starting list; the app gives you the current one.
Finding the words your audience already uses
Your comments are the keyword research
The most accurate list of phrases your audience uses is already in your comment section and your direct messages. Somebody asks \"how do I stop the starter from smelling like nail polish\" and that sentence is a video title, a hook and a caption line, in the exact words the next person will type. Collect ten of those before running any tool, and the output becomes specific instead of plausible.
Specific beats short
A short phrase is competitive and vague. A long phrase is nearly free and self-selecting: the person who types six words is not browsing, they are looking for the answer to one problem, and they will watch a video that starts with their own sentence. On social platforms the long phrase is also what the search box suggests, because the engine is matching whole queries, not words.
The caption is a title, not a caption
The first line of a caption is the only line shown before \"more\", and it is what the platform reads first. Put the phrase there, in a sentence a person would say. The same phrase goes into the on-screen text of the first two seconds. Three placements, one phrase, one video.
Hashtags: fewer, and honest
Three to ten tags that describe what the video is about. Not #fyp, not twenty tags from a list of \"best hashtags for growth\". A tag that does not describe the video misleads the classification and wastes the caption's last line, which is prime space for a call to action.
Give the account a topic before chasing reach
A new account with no topic gets shown to a random sample, and the sample does not watch. Five videos about the same narrow thing teach the platform who to show the sixth to. The wide layer of keywords is for reach and works only once that teaching has happened - which is why the mix matters more than any single phrase.
Watch the saves, not the likes
A save is somebody deciding to come back, and it is the strongest signal that a video answered a question rather than filled a minute. Videos built on the question layer get fewer views and more saves; both numbers are useful, but only one of them compounds.
Refresh the list every quarter
Phrases age. A term that was specific a year ago is now the generic one everybody uses, and the search suggestions have moved on. Re-run the tool with the last quarter's best video as the description: the words that come back are the ones the audience is using now, and the difference between the two lists is your next content plan.
Five runs a day covers a quarter's worth of posts if you plan them together. The list is not the deliverable - the calendar made from it is.
Questions about keywords
Does it show search volume?expand_more
How is this different from a hashtag tool?expand_more
Can I use the same list for every platform?expand_more
Why are there questions I have never seen?expand_more
What are the gaps for?expand_more
Find the words before you film
Describe the account, take the phrases and questions, and build next month's posts from the list.
fact_checkQuick facts
- A keyword list is a hypothesis until it is checked against the search volume of the platform you publish on.
- Long phrases with a clear intent convert better than one broad word, because the person typing them already knows what they want.
- A language model reads text in tokens, and one token is roughly four characters of English, so a 300-word answer is about 400 tokens.menu_bookOpenAI Cookbook
- The output depends on the prompt: the same request with a different tone or length instruction produces a different text, not a different answer quality.menu_bookOpenAI Cookbook
- Google builds a title link and a snippet from the page itself, and it may use a heading or a sentence from the body instead of the meta description.menu_bookGoogle Search Central
helpQuestions people ask
How do I find keywords for social media?expand_more
What is the difference between a keyword and a hashtag?expand_more
Are AI keyword suggestions accurate?expand_more
hubWhat this page is about
- Large language modelopen_in_new
A model that predicts the next token of a text; the site sends the prompt to a hosted model and returns its answer.
- Answer engine optimisationopen_in_new
Writing a page so that a search engine or an AI assistant can lift one clear, sourced answer out of it instead of guessing.
library_booksOfficial sources cited on this page
- How to count tokens with tiktokenOpenAI Cookbook
- How Google generates titles and snippetsGoogle Search Central
- FAQPage structured dataGoogle Search Central
- HowTo structured dataGoogle Search Central
Keep going with the next tool
The tools in this group that do the next piece of the same work, all switched on right now.
Switched-off tools are not listed here: only what works right now.