The most consequential change in social discovery over the last few years has nothing to do with hashtags. It is that platforms now read your actual words — captions, on-screen text, transcribed speech — and match them against what people search for.

Keyword search arrived quietly

Instagram search returns posts matching caption text, not just tags. TikTok search is used heavily as a general reference tool. YouTube has always been keyword-driven. Pinterest was built that way. The tag-only era ended without an announcement.

What this means for captions

The first sentence of your caption is now doing search work. "Obsessed 😍" tells the platform nothing. "Three ways to fix a wobbly kitchen chair with one tool" tells it a great deal, and matches what someone would actually type.

This is not a licence to stuff keywords. It is an argument for writing plainly about what the post contains, in the first line, before the personality kicks in.

Speech and on-screen text count

Short-form video platforms transcribe audio and read on-screen text. That means your spoken opening line is a ranking signal. Saying the topic out loud in the first three seconds serves the viewer and the classifier simultaneously.

Where tags still fit

Tags remain the cleanest way to say "this belongs to this category," and they cover query variations your caption could not naturally include. A caption cannot gracefully contain six phrasings of the same idea; six tags can.

Think of it as a division of labour. Caption carries the primary phrasing in natural language. Tags carry the variants, the community labels and the categorisation.

A practical test

Before posting, ask: if someone typed the thing this post is about into the search bar, would the words in my caption match it? If not, rewrite the first line. That single habit does more for discovery than any tag adjustment.

Alt text is free

Image alt text is read by both accessibility tools and classification systems, and almost nobody fills it in. Writing a plain description of the image takes fifteen seconds and helps twice.

Put this into practice — the generator applies everything above automatically. Build a set now.