How our tools work
This page describes, without marketing language, exactly what happens between typing a topic and seeing a list of keywords. It is written so that you can judge for yourself how much weight to give the output.
- 1
You describe the topic
You write a short phrase describing your content, for example beginner sourdough bread. That text is the only input. We do not look at your account, your history or any social profile.
- 2
The request goes to our server
Your browser sends the topic to our own backend. This matters for privacy: no third-party script in your browser reads what you typed, and no API key is exposed on the client side.
- 3
A platform-specific instruction set is added
Each generator adds instructions describing how discovery works on that platform: how many tags are reasonable, whether broad or niche terms should dominate, what to avoid. The same topic sent to the Instagram and YouTube generators produces deliberately different output.
- 4
A language model generates the list
The model returns a structured list of keywords or hashtags. We use models from Google's Gemini family. The model reasons about your topic and the platform; it does not read another creator's tags and it does not query a scraped database of hashtags.
- 5
The list is returned to your browser
You see the list with a copy button. We do not require an account, so nothing about the request is attached to your identity, and we do not build a personal search profile from the topics you enter.
What the output is, and what it is not
The output is a reasoned suggestion list. It is not a search volume report, it does not include keyword difficulty scores and it does not contain paid search data. If you need measured volumes or competition metrics, you need a data-driven keyword platform; those are a different category of tool.
What this generator gives you is speed and coverage: a starting set on a topic you may know nothing about yet, in seconds, for free. That is most of the value when you are planning a week of posts rather than a year of strategy.
Why results vary between runs
Language models are probabilistic. Two runs on the same topic will usually return broadly similar lists with differences in ordering and a few different entries. That is a feature rather than a bug: it gives you alternative angles when you return to a topic you have already covered.
It also means the output should not be treated as a fixed canonical set. Two creators in the same niche will get slightly different lists, and that is closer to how keyword research works in practice than a single immutable table would be.
How to judge a generated list
Ask three questions of every line. Does it describe my content honestly? Would someone actually search for this phrase? Could I compete for it, or is it owned by accounts far larger than mine? Anything that fails all three should be deleted before you paste it anywhere.
The strongest signal that a list is working is that it contains a term you had not thought of but recognise immediately as something your audience says.
Supporting content around each tool
Every tool page carries written guidance: how that platform handles discovery, what a sensible number of hashtags looks like, and the errors that hurt results most often. The blog goes deeper, covering platform changes and the reasoning behind the advice.
That content exists because a generator without context invites misuse. A list of thirty hashtags dumped into a caption is worse than five chosen with intent. The guides exist to make the intent part easier.