PostCrows
TikTok indexes three separate layers for search: your caption copy, your on-screen text through optical character recognition, and your spoken audio through speech-to-text covering 30-plus languages. Each is a distinct ranking input. A post that places its primary keyword in all three layers of the keyword stack gets a substantially stronger intent match than one relying on hashtags alone.
Most creators optimize one layer - usually the caption - then wonder why weaker content outranks them in search. The answer is coverage. Three confirmations of the same topic beat one.
The three layers TikTok indexes
| Layer | How it is read | Where the keyword goes | Works on carousels? |
|---|---|---|---|
| Caption | Direct text parsing | First line, naturally phrased | Yes |
| On-screen text | OCR on frames and slides | Opening frame or slide 1 | Yes - strongest layer |
| Spoken audio | Speech-to-text, 30+ languages | First few seconds aloud | Only with voiceover |
Caption
The cheapest layer, and the one most people use badly. Lead with the keyword in a real sentence rather than burying it after three lines of setup. “How to schedule TikTok carousels when the native scheduler won’t” classifies cleanly. “New post 🔥 #fyp #viral” classifies as nothing. See caption formulas for structures that carry keywords without sounding robotic.
On-screen text
TikTok runs OCR across frames, so words rendered on your slides are indexable content, not decoration. This is why text-heavy carousels punch above their weight in search. Put the primary keyword in the visible text of slide one and keep it legible - thin type over busy imagery fails OCR the same way it fails human readers. The on-screen text guide covers sizing and contrast.
Spoken audio
Speech recognition transcribes what you say and indexes it, so say the keyword out loud in the first few seconds in a natural sentence. This layer is where video has a structural advantage over photo posts, and it is one of the few real arguments for adding AI voiceover to faceless content.
The takeaway: caption, OCR text, and transcribed audio are three separate inputs - treat them as three separate jobs.
How do you stack a keyword across all three layers?
Take one primary keyword and place it deliberately:
- Say it within the first three seconds. Early placement confirms topic intent before viewers drop off.
- Show it in the opening frame or slide 1. The same words, not a clever paraphrase.
- Open the caption with it. First line, natural phrasing, no keyword list.
- Support it with 2-3 close variants spread across body slides and the rest of the caption.
- Add one question form somewhere - “how many slides should a TikTok carousel have” style - because a large share of search traffic is phrased as questions.
The structure worth internalizing is 1 primary keyword + 2-3 supporting variants + 1 question form per post. Stuffing beyond that dilutes classification instead of strengthening it: if TikTok reads your post as four unrelated things, it tests it on the wrong audience and distribution stalls before content quality matters.
The takeaway: one primary keyword in all three layers, plus 2-3 variants and one question form - nothing more.
Does the keyword stack work for photo carousels?
Yes, with one caveat. Carousels have no spoken audio unless you add a voiceover, so you work with two of the three layers. That is rarely a disadvantage, because the OCR layer is unusually strong here: a 7-slide educational deck may hold 60 to 100 clearly rendered words, all indexable, versus a video where on-screen text is sparse and transient.
Practical adjustments:
- Slide 1 carries the keyword in its headline text, doing double duty as hook and search signal.
- Body slides reinforce variants naturally, since you are explaining the topic anyway.
- Keep contrast high. OCR and human readability fail on the same slides. Design at 1080x1920 with critical text away from the top 10% and bottom 20%.
- Add a voiceover only if it serves the viewer. Recording one purely for the audio layer rarely pays for itself on a photo post.
The takeaway: carousels lose the audio layer but their OCR layer is unusually rich - lean into slide text.
Which keywords are worth stacking?
Stacking the wrong keyword perfectly still gets you nothing. Two free sources fix that:
- Creator Search Insights, in-app - search “creator search insights” to open it. Its Content gap filter surfaces queries with high demand and few good answers, which is the fastest ranking path for a small account. A Searches by followers filter unlocks at 1,000+ followers. Full walkthrough in the Search Insights tool guide.
- Search analytics on your own posts, showing which queries actually surfaced your content. That is the feedback loop telling you whether the stack landed. Surfacing for terms you never targeted means your classification is drifting - see keyword metadata.
One constraint: new accounts rarely rank in search immediately, because the algorithm has not classified your topic yet. Plan roughly 80% broad discovery content and 20% search-targeted content early, shifting toward 30-40% search content by week four.
The takeaway: find gap queries first, then stack them - and give a new account four weeks before judging search results.
The bottom line
TikTok’s three-layer keyword stack is the highest-leverage SEO move available because it costs nothing extra and almost nobody completes it. Pick one primary keyword from a Creator Search Insights content gap, say it aloud in the first three seconds, render it as legible on-screen text in the opening frame or slide one, and lead the caption with it in a natural sentence. Support it with two or three close variants and one question form, then stop - over-stuffing dilutes classification and misdirects your test audience. Carousels work with two layers instead of three, but their dense slide text makes OCR unusually strong. Check search analytics after a week to see which queries surfaced you, and build the next post from that. Broader context in TikTok SEO.