A swipe file is a set to copy from. This one is 86 AI product explainer videos from 69 companies, uploaded between 5 January 2026 and 3 August 2026, taken apart on eight build patterns: 70 open on something other than the product, 52 name or show it by 0:10, 79 put an interface on screen, 55 carry no screen capture, 62 use a voiceover, 66 hold one ground, 56 run 0:46 to 2:00 and 65 close on the product's name. Stills from Orange Business, Cequence Security, Prompt Security, Campfire, etc. illustrate the patterns.
188 product videos carrying a full visual record were screened against a rule written before screening; 86 qualified, from 69 companies. Every one explains what an AI product does and how it does it, runs 0:20 to 4:00, and was uploaded to YouTube between 5 January 2026 and 3 August 2026. Launch teasers, brand spots, tutorials, testimonials and keynotes were excluded. In this report a video is one AI product explainer video, a company is the one that sells the product, and a pattern is a coded choice counted across the whole set; no period is compared with another.
ADVIDS Research · 86 AI product explainer videos coded · 69 companies · 774 sampled frames · 909 timed segments · published 14 September 2026 · 21 min read
Top takeaways
70 of 86 AI explainers open on something other than the product
70 of 86 AI product explainer videos uploaded in 2026 open on something other than the product interface, led by a line of type in 29 and people or a scene in 19; 16 open on the interface, Blink, Superblocks and Campfire among them.
52 of 86 name or show the product inside ten seconds
52 of 86 AI product explainer videos from 2026 name or show the product by 0:10, at a median 0:07; 9 wait past 0:30 or never do, including Akto, Amplitude, Doxis, HCLTech, etc.
79 of 86 put a product interface on screen; dashboards lead
A product interface appears in 79 of 86 AI product explainer videos from 2026, a dashboard in 63 and a chat panel in 43; the 7 without one include Salesforce, Atlassian, Lyzr, Pinecone, etc.
55 of 86 carry no screen capture of the product
55 of 86 AI product explainer videos from 2026 carry no screen capture of the product and 2D animation is listed in 50; capture appears in 31, Savant Labs, Cequence Security and Superblocks among them.
Voiceover carries 62 of 86; kinetic type appears in 49
A voiceover carries 62 of 86 AI product explainer videos from 2026; 24 run on music alone, all 5 from Anthropic among them, and captions are burned in for 9, including Alloy, Cegid, Cequence Security, Kantar, etc.
66 of 86 hold one ground; light leads dark 41 to 25
66 of 86 AI product explainer videos from 2026 hold one ground, light in 41 and dark in 25; 20 mix light and dark frames, Ziflow among them.
56 of 86 run between 0:46 and 2:00
56 of 86 AI product explainer videos from 2026 run 0:46 to 2:00, at a median 1:19; 13 run 0:45 or less and 17 run past 2:00, up to 2:45 from Cequence Security.
65 of 86 AI explainers close on the product's name
65 of 86 AI product explainer videos from 2026 close on the product's name, 59 on a logo card; 24 carry a URL or an ask in the closing text, IPRally among them.
Questions this report answers
How do AI product explainer videos open?
Most open away from the product: in this 2026 set a line of type comes first in 29 of 86 videos and people or a scene in 19. Only 16 open straight on the interface, Blink and Superblocks among them.
When does an AI explainer video first show the product?
At a median 0:07, and by 0:10 in 52 of 86 videos in this set; at that moment 39 show the interface and 20 a card naming the product.
Do AI explainer videos use screen recordings or animation?
Animation more often: 2D animation is listed in 50 of 86 videos and an animated interface mockup in 40, against screen capture in 31. Savant Labs runs 2:36 on capture alone.
Do AI product videos need a voiceover?
Most use one: 62 of 86 in this 2026 set, while 24 run on music alone, all 5 Anthropic videos among them. Kinetic-type segments appear in 17 of the 24 music-only videos.
How long is an AI product explainer video?
The median runs 1:19, with the middle half between 0:54 to 1:49; voiceover videos run a median 1:32 and music-only videos 0:59. The longest, from Cequence Security, runs 2:45.
How was this swipe file built?
ADVIDS Research screened 188 visually coded videos against a rule written before screening; 86 from 69 companies qualified, and 10 launch teasers and brand spots were among the 102 excluded. Each video was coded from nine sampled frames and 909 timed segments.
How to read this report
The set is 86 AI product explainer videos from 69 companies, uploaded 5 January 2026 to 3 August 2026; each of the eight patterns is a coded choice counted across it, in the order a video plays.
A share such as 70 of 86 means 70 of the 86 videos carry the pattern. The base is all 86 unless a sentence names a smaller one, such as the 60 voiceover scripts of 30 words or more or the 79 videos that show an interface.
Timecodes are seconds from the first frame; a median is the middle video among those that carry the measure. A pattern gets a chapter only when at least half its base carries it.
Pattern 01 · Opening move · 86 AI product explainer videos, 5 January 2026 to 3 August 2026
1. 70 of 86 AI explainers open on something other than the product
70 of 86 AI product explainer videos uploaded between 5 January 2026 and 3 August 2026 open on something other than the product interface. The first image after any blank lead-in is a line of type in 29, people or a scene in 19, the product's name in 12 and a graphic or diagram in 10; 16 open straight on the interface. 32 start on a blank card, and their first image lands at a median 0:04. Among the 60 voiceover scripts of 30 words or more, the first sentence is a claim in 31, names the product in 14, states a problem in 11 and asks a question or a what-if in 4. The screen waits for a sentence.
The 16 that open on the interface include Anthropic, Workday, Blink, Campfire, etc. In the still, Orange Business opens on its problem line, set in kinetic type over an illustrated contact centre.
Opens on something other than the interface, 70 of 86. Companies in this group include OpenAI, SAP, Bloomfire, HPE, etc.
Blink goes the other way. Its January 2026 Agent Builder video opens at 0:00 on the product's own create prompt and builds a landing page before a single line of type appears. Where the product is a prompt box, the prompt box is the opening.
Source: ADVIDS Research coded set, 86 videos from 69 companies, uploaded 5 January 2026 to 3 August 2026
The first image after any blank lead-in card, across all 86 videos. The interface is third.
Pattern 02 · First named or shown · 86 AI product explainer videos, 5 January 2026 to 3 August 2026
2. 52 of 86 name or show the product inside ten seconds
The product is named or shown by 0:10 in 52 of 86 AI product explainer videos from 2026, and by 0:05 in 33; the median first moment is 0:07, with the middle half between 0:00 to 0:16. At that second the screen carries the product interface in 39 and a card or line naming the product in 20, 59 in all; in the other 27 it shows type without the name, people, a graphic or a blank card. Of the 77 videos that both name and show the product, 40 name it first, 30 show the interface first and 7 do both in the same segment. The name is on screen no later than it is spoken in 48 of the 83 that name it. Name and screen arrive inside the opening, not after it.
The 9 that wait past 0:30 or never name or show the product include Kroolo, PNY, Akto, etc. In the still, Cequence Security puts its AI Gateway name card on screen at 0:00 and keeps it there while the voiceover starts.
Named or shown by 0:10, 52 of 86. Companies in this group include Amplience, AppsFlyer, Arm, Asana, etc.
Kantar goes the other way. Its April 2026 BrandDigital video runs particle fields and search-bar lines for 0:56 before a card names the product, and the dashboard follows at 1:04. The first minute belongs to the problem the product measures.
Source: ADVIDS Research coded set, 86 videos from 69 companies, uploaded 5 January 2026 to 3 August 2026
The first second at which the name is on screen or spoken, or an interface appears, whichever comes first.
Pattern 03 · Surface shown · 86 AI product explainer videos, 5 January 2026 to 3 August 2026
3. 79 of 86 put a product interface on screen; dashboards lead
A product interface appears in 79 of 86 AI product explainer videos from 2026. Dashboards appear in 63, chat panels in 43, screen-capture passages in 14, notifications in 10, terminals in 7 and device frames in 7. The first interface lands at a median 0:13, and interfaces cover a median 48 per cent of runtime in the videos that show one, half or more in 33 of 79. The 7 with no interface carry the product as type, diagrams or hardware. The AI product is shown as a dashboard more often than as a conversation.
The 7 without an interface include Salesforce, Atlassian, Doxis, etc. In the still, Prompt Security opens its red-teaming video on the Start a New Scan screen at 0:03.
Product interface on screen, 79 of 86. Companies in this group include IBM, Sword Health, Airia, Aviso, etc.
Pinecone goes the other way. Its February 2026 explainer runs 2:27 of diagrams, embeddings flowing into a vector database and back as query results, on a near-black ground, and never puts an interface on screen. Where the product is infrastructure, the architecture diagram is the product shot.
Source: ADVIDS Research coded set, 86 videos from 69 companies, uploaded 5 January 2026 to 3 August 2026
A video can show several surfaces; each bar counts the videos with at least one segment of that surface.
Pattern 04 · Build and technique · 86 AI product explainer videos, 5 January 2026 to 3 August 2026
4. 55 of 86 carry no screen capture of the product
55 of 86 AI product explainer videos from 2026 carry no screen capture of the product in their coded build. 2D animation is listed in 50, an animated interface mockup in 40, kinetic type in 20, live action in 18, stock in 10 and 3D in 8; screen capture is listed in 31. Of the 79 that show an interface, 48 draw or mock it rather than capture it. Build codes are medium confidence for 22 of the 86, including 8 of the 31 capture codes and 5 of the 8 3D codes. The interface in an AI explainer is drawn more often than it is recorded.
Companies with 3D in the build include Dell Technologies, ElevenLabs, GitLab, Infobip, etc. In the still, Campfire configures an accounting agent in an animated mockup, and its build lists no screen capture.
No screen capture in the build, 55 of 86. Companies in this group include Board, Celigo, ClayHog, Crisp, etc.
Savant Labs goes the other way. Its January 2026 reconciliation video runs 2:36 with screen capture as its only coded build, walking a flow canvas from bank-statement PDFs to journal entries. Where the product's case is a sequence of real steps, the capture carries it.
Source: ADVIDS Research coded set, 86 videos from 69 companies, uploaded 5 January 2026 to 3 August 2026
Build codes are listed per video, several each; a bar counts the videos whose codes include that method.
Pattern 05 · Delivery · 86 AI product explainer videos, 5 January 2026 to 3 August 2026
5. Voiceover carries 62 of 86; kinetic type appears in 49
A voiceover carries 62 of 86 AI product explainer videos from 2026; the other 24 run on music with no voiceover. Kinetic-type segments appear in 49 of 86, and in 17 of the 24 music-only videos. Captions are burned into the picture in 9 and a presenter appears on camera in 7. Voiceover scripts run a median 196 words. All 5 Anthropic videos in the set are music-only; without Anthropic the voiceover share moves 4.5 points, under the floor. Where there is no voice, the type is the script.
The 9 with burned-in captions include Ofelia, Orange Business, Otio, Savant Labs, etc. In the still, Alloy sets context matters in kinetic type with the voiceover burned in as a caption beneath it.
Voiceover, 62 of 86. Companies in this group include Contentstack, Epicor, Genpact, Google Cloud, etc.
Board goes the other way. Its April 2026 planning-platform video runs 1:00 with no spoken word and no burned caption; mockups of news, weather and finance widgets and lines of kinetic type carry every claim.
Source: ADVIDS Research coded set, 86 videos from 69 companies, uploaded 5 January 2026 to 3 August 2026
Pattern 06 · Frame · 86 AI product explainer videos, 5 January 2026 to 3 August 2026
6. 66 of 86 hold one ground; light leads dark 41 to 25
66 of 86 AI product explainer videos from 2026 hold one ground across their nine sampled frames: light in 41 and dark in 25; the other 20 mix light and dark frames. Median frame brightness is 140 on a 0 to 255 scale, 191 for light videos and 58 for dark ones, and 47 of 86 average above 128. The measure is luminance across every sampled frame, title cards and people included, not the product's own theme setting. One ground runs through three in four of these videos, and light is the more common one.
The 20 mixed videos include Nexthink, Rackspace Technology, Workiva, Yext, etc. In the still, Microsoft holds a near-white ground from its chat prompt through its task board.
One ground, light or dark, 66 of 86. Companies in this group include Dext, IPRally, LangChain, Make, etc.
Ziflow goes the other way. Its May 2026 ReviewAI demo cuts between bright package mockups and a dark review interface, a mixed video at a mean brightness of 94. Where the product reviews other teams' creative, the creative sets the light and the tool frames it.
Source: ADVIDS Research coded set, 86 videos from 69 companies, uploaded 5 January 2026 to 3 August 2026
Luminance across the nine sampled frames of each video, not the product's theme setting.
Pattern 07 · Runtime · 86 AI product explainer videos, 5 January 2026 to 3 August 2026
7. 56 of 86 run between 0:46 and 2:00
56 of 86 AI product explainer videos from 2026 run between 0:46 and 2:00. The median runs 1:19, the middle half spans 0:54 to 1:49, and the set ranges from 0:30 to 2:45. 13 run 0:45 or less, 16 run 0:46 to 1:00, 20 run 1:01 to 1:30, 20 run 1:31 to 2:00 and 17 run past 2:00. Voiceover videos run a median 1:32 and music-only videos 0:59. Videos with a voice run half a minute longer than videos without one.
The 13 running 0:45 or less include Slack, Anthropic, Bloomfire, etc.; the 17 running past 2:00 include ScienceLogic, Superblocks, Zoom, etc. In the still, Genpact explains Autonomous IT in 0:58, with a node diagram of its agent just before the logo card.
Runs 0:46 to 2:00, 56 of 86. Companies in this group include Microsoft, Neo4j, Nooks, Perplexity, etc.
Cegid goes the other way. Its July 2026 Retail Live Store video runs 0:30, the shortest length in the set alongside Epicor and Slack, and still fits a voiceover of 64 words, a store scene and the app's product suggestions.
Source: ADVIDS Research coded set, 86 videos from 69 companies, uploaded 5 January 2026 to 3 August 2026
Runtime read from each file, in five fixed ranges chosen before counting.
Pattern 08 · Close · 86 AI product explainer videos, 5 January 2026 to 3 August 2026
8. 65 of 86 AI explainers close on the product's name
65 of 86 AI product explainer videos from 2026 close on the product's name: a logo card in 59 and a title card naming the product in 6. 12 close on a line of type without the name and 9 on an interface, a person or a graphic. The on-screen text coded in the last two segments carries a URL or an ask, such as book a demo or start now, in 24 of 86, and a URL in 16. The last frame names the product; fewer than a third say where to go next.
The 12 that close on a line of type include StackAI, Amplience, Amplitude, etc. In the still, IPRally ends on its name, a two-word line and its URL.
Closes on the product's name, 65 of 86. Companies in this group include Prompt Security, Rillion, Zeta Global, Ziflow, etc.
Contentstack goes the other way. Its April 2026 AXP overview ends on a lineup of feature icons for agents and automation, CDP, personalisation and headless content, not on a logo or a name card.
Source: ADVIDS Research coded set, 86 videos from 69 companies, uploaded 5 January 2026 to 3 August 2026
The last segment that is not a blank card, across all 86 videos.
Benchmarks
Descriptive figures for placing your own AI product explainer against the set.
Length, timing and picture
| Measure | Value | What it means |
|---|---|---|
| Runtime, median | 1:19 | Half the 86 videos run shorter than this |
| Runtime, middle half | 0:54 to 1:49 | The span of the middle 50 per cent of videos |
| First named or shown, median | 0:07 | Among the 85 videos that name or show the product |
| First named or shown, middle half | 0:00 to 0:16 | The span of the middle 50 per cent of those videos |
| First named, median | 0:13 | Among the 83 videos that name the product on screen or aloud |
| First interface on screen, median | 0:13 | Among the 79 videos that show an interface |
| Interface coverage, median | 48 per cent | Share of runtime with an interface on screen, among those 79 |
| Words spoken, median | 196 | Among the 62 voiceover videos; middle half 124 to 256 |
| Frame brightness, median | 140 | Mean luminance across nine sampled frames, 0 to 255; middle half 85 to 190 |
| Timed segments per video, median | 10 | Coded motif segments, a proxy for scene changes |
What fewer than half do
| Pattern | Videos | Base | Companies include |
|---|---|---|---|
| Screen capture of the product in the build | 31 | 86 | Anthropic, OpenAI, Airia, etc. |
| URL or ask in the closing text | 24 | 86 | Anthropic, Akto, Amplience, etc. |
| Mixed light and dark frames | 20 | 86 | Airia, Bloomfire, Cegid, etc. |
| Live action in the build | 18 | 86 | Bloomfire, SAP, Amplitude, etc. |
| Opens on the product interface | 16 | 86 | Anthropic, Workday, Blink, etc. |
Each is carried by fewer than half its base, so none gets a chapter; the five closest to half are listed. Screen capture (31 of 86) and a URL or ask on the end card (24 of 86) are the two a brief is most likely to assume; both are minority choices in this set.
What to do with this
- Pattern 01 If 70 of 86 open away from the product, write the first line of type or the first scene into the brief as deliberately as the first screen.
- Pattern 02 If 52 of 86 name or show the product by 0:10, mark on the storyboard the frame where the name first appears, and put the interface or a name card on it.
- Pattern 03 If 63 of 86 show a dashboard and 43 a chat panel, decide which screen the demo environment must have populated before production starts.
- Pattern 04 If 55 of 86 carry no screen capture, decide early whether the interface is animated from design files or recorded from a live tenant; the two need different inputs.
- Pattern 05 If 24 of 86 run without a voiceover and 17 of those use kinetic type, write the on-screen lines as the script whenever there is no voice.
- Pattern 06 If 66 of 86 hold one ground, choose light or dark in the creative brief and hold the graphics to the product's own theme.
- Pattern 07 If 56 of 86 run 0:46 to 2:00, set the length in the brief before the script is written rather than cutting down afterwards.
- Pattern 08 If only 24 of 86 carry a URL or an ask in the closing text, decide what the end card asks for, not only which logo it shows.
Methodology
| Vertical | Videos |
|---|---|
| devtools | 13 |
| fintech | 8 |
| productivity | 5 |
| ai | 4 |
| cx | 4 |
| martech | 4 |
| enterprise | 4 |
| data | 3 |
| other verticals | 41 |
| Coding scheme | Values | Definition |
|---|---|---|
| Opening move | interface · type · people · name · graphic · blank lead-in | Coded per timed segment: the motif of the first segment that is not a blank card. A blank card is a title card or background with no on-screen text coded. Type is a kinetic-type or title segment without the product's name; name is a logo card or a card that names the product. |
| First spoken line | claim · name · problem · question or what-if | Transcribed: the first sentence of every voiceover script of 30 words or more, sorted by a published word rule. Lyrics transcribed as captions are left out. |
| First named or shown | seconds · what is on screen then | The earliest of three moments: the product's name in on-screen text or a logo card, the name in the captions, and the first interface segment. The screen at that second is classed as for the opening move. |
| Surface | dashboard_ui · chat_ui · screen_recording_generic · notification_alert · terminal_ui · handheld_device_ui | Coded per timed segment, with the first timecode and the share of runtime covered. |
| Build | 2d_animation · ui_mockup · screen_recording · kinetic_type · live_action · stock · 3d_cgi | Coded from frames: the production methods visible, several per video, each video's codes carrying a confidence of high or medium. screen_recording is capture of the product; ui_mockup is an animated recreation of it. |
| Delivery | voiceover · music_only · captions burned in · kinetic-type segments | Voiceover or music only from audio analysis; burned-in captions coded from frames; word counts from auto or manual captions. |
| Frame | light · dark · mixed · brightness 0 to 255 | Computed, not coded: mean luminance across the nine sampled frames, sorted light, dark or mixed. It covers every frame, not the product interface alone. |
| Runtime | seconds | Read from the file. |
| Close | name · type · interface · people · graphic · URL or ask | The motif of the last segment that is not a blank card, classed as for the opening move, and whether the on-screen text coded in the last two segments carries a URL or an ask. |
| Not coded | altitude · cast | What each video argues and who appears as whom. The library records neither, and this report does not code them from transcripts. |
Cite this report
Jai Ghosh, Advids. The 2026 AI Product Video Swipe File: 86 Explainer Videos Deconstructed. 14 September 2026. https://advids.co/blog/ai-product-explainer-video-swipe-file
About Advids
Advids produces video for AI, SaaS and DeepTech companies. ADVIDS Research publishes coded studies of what those companies ship, with the dataset attached so every count can be checked.
Image credits and licensing
Video frames on this page are reproduced for analysis and commentary. Copyright in each frame remains with the company credited in its caption; to license a frame, contact that rights holder. Advids grants no rights over them.
Charts and images credited ADVIDS Research, and the coded dataset behind them, are © 2026 Advids and licensed CC BY 4.0: reuse them with attribution to Advids and a link to this page.