TikTok and Instagram Now Work Like Search Engines. Are You Optimising For That?

The short answer

Yes. TikTok, Instagram and YouTube all now run two separate discovery systems in parallel: a recommendation feed that guesses what you'll engage with, and a search index that matches what people type or say. Most creators only write for the first one. If your captions, spoken words and on-screen text never state your topic in plain language, you are invisible to the second — and search traffic tends to be steadier and more intentional than feed traffic, because the person typing already wants what you're offering.

Two different jobs, one video

It's worth being precise about what's actually happening, because "the algorithm" has become a catch-all term that hides a real distinction.

The For You feed is a recommendation system. It watches what you've engaged with before and tries to predict what will hold your attention next. It rewards watch time, rewatches, shares and comments, and it doesn't much care what your video is about — it cares whether people like you tend to finish videos like yours.

Search is a different problem entirely. Someone has typed or spoken a query with a specific intent — "how to fix a squeaky bike chain", "best budget tripod for phones" — and the platform is trying to match that intent to content that actually answers it. This is much closer to how Google has always worked than to how a For You feed works.

The reason this matters now is that TikTok and Instagram have been investing hard in the second system. TikTok has shipped Creator Search Insights, a tool that shows what your audience is searching for and, more usefully, where there's search demand with very little content answering it. Instagram has quietly expanded keyword search across captions, alt text and audio for a couple of years now. Neither company advertises this loudly, because "we're becoming a search engine" undercuts the short-form entertainment pitch. But functionally, that's what's happening.

Why this catches repurposed content out specifically

If you clip long-form content into shorts, this is where you're most exposed. A clip pulled from the middle of a podcast or a livestream is often captioned for the hook — "you won't believe what happened next," a teaser, an in-joke that made sense in context. That caption is written to win the feed. It states nothing about the topic, so it gives the search index nothing to match against.

The fix isn't to abandon the hook. It's to accept that a clip now needs to do two jobs at once: earn attention in the first second for the feed, and state its actual subject somewhere — caption, on-screen text, or the words you say out loud — for search. Those aren't the same sentence, and trying to force one line to do both usually produces something that's mediocre at each.

What we'd actually change in a caption

• Say the topic out loud, early. All three platforms transcribe spoken audio and index it. A video where the topic is only implied visually, with a vague voiceover, gives the transcript nothing to work with. If you're demonstrating something, name it in the first few seconds — not just show it.
• Put the actual query phrase in on-screen text. Text overlays get read by OCR and indexed too. If someone would plausibly type "how do I remove a stripped screw," having those words appear on screen, once, in plain language, does more for search than five hashtags.
• Write a second caption line that a search engine can parse. Keep your hook as the opening line, then follow it with a plain-language sentence describing what the video actually covers. You lose nothing on the feed side — most people never read past the hook anyway — and you gain a sentence that search can actually match against.

Where hashtags fit, honestly

Hashtags still do something, but treat them as topic and community signals rather than a keyword-stuffing exercise. Stacking ten loosely related tags at the end of a caption reads as spam to a human and doesn't reliably move you up a search result the way exact-match keywords used to on early-2010s Google. The platforms increasingly use semantic matching — understanding that "budget tripod" and "cheap phone stand for filming" mean similar things — rather than rewarding you for typing the same word five times. A couple of specific, accurate hashtags beat a wall of generic ones.

We'll say plainly what we don't know here: none of the platforms publish how spoken transcription, on-screen text and caption text are weighted against each other, and that weighting almost certainly shifts without announcement. Anyone who tells you the exact ratio is guessing. The safe move is to cover all three rather than bet on one.

What this doesn't replace

None of this substitutes for a video actually being good. Search can get someone to click; it can't make them stay, and watch time still decides whether that video then gets pushed further by the recommendation side. Treat search optimisation as a way to be findable by people who are already looking for what you make — not a shortcut around making something worth finding.

It's also not worth over-engineering. You don't need a keyword spreadsheet for every post. The habit that matters is small and repeatable: before you publish, ask whether someone with the right intent, typing a plain-language search, would ever land on this video. If the honest answer is no, add one sentence that fixes it.

Where this fits

This is exactly the kind of detail that gets lost when a long recording is being cut into a dozen clips under time pressure — the hook survives, the plain-language description doesn't. CORE handles repurposing and channel management day to day, and building a search-friendly caption pass into that workflow is a small, permanent fix rather than something to remember on every upload.