# LinkedIn ad formats for B2B: choose the container for the argument > Choose the format by the shape of the argument: one claim with one proof fits a single image, steps in order fit a carousel, a reference fits a document, a person or a product in motion fits a video, and a view that needs an owner fits a thought leader post. Public benchmarks cannot separate the format from the offer, because advertisers pick the format to suit the offer. To find out in your own account, test two formats at a time with the claim, proof and offer held fixed, for at least 14 days in LinkedIn's A/B test. Source: https://nimblet.io/guides/linkedin-ad-formats. Updated 2026-10-23. We read the LinkedIn pages and the ad library entries below on 22 and 23 October 2026. LinkedIn can change its specs and its objective rules, so check Campaign Manager before you brief a designer. The ads are real and public. A number inside an ad is the advertiser's own claim. A sentence that says "our reading" is our judgement, not a measured fact. In this guide, "format" means any LinkedIn ad format. "Container" means one of the five formats we compare: single image, carousel, document, video and thought leader post. ## Does the format or the concept decide the result? Nobody outside your own account can tell. Inside your account, you can tell only when the concept stays the same across formats. A format changes what fits in the ad, how much attention the buyer must pay and what the buyer does next. So a change of format also changes the argument, unless you carry the claim, the proof and the offer across on purpose. Our reading starts with the buyer. A B2B buyer spends attention and risks looking foolish in front of colleagues. Each container quotes a price for that attention before the buyer reads a word. A single image costs a glance. A carousel costs one swipe per card. A document costs pages of reading. A video plays at its own pace, whether the buyer wants that or not. So the container is part of the promise, and a container that asks for more than the argument is worth loses the buyer early. The ad library shows how one advertiser spreads one concept over several containers. Freshworks promotes its Customer Service Benchmark Report '26 in a [74-page document ad](/guides/ads/1597821666.jpg), a message ad and a single image ad, and the library holds many more of its ads for the same report. The three below show in the Germany filter and ran between 3 September and 23 October 2026. One report figure leads all three: - Single image ad: ["Last year it was 32%."](/guides/ads/1619720746.jpg) - Message ad: ["A year ago, the number was 32%. Today, 43.6% of customer service teams use AI across their service workflows."](/guides/ads/1589694386.jpg) - Document ad: ["43.6% now use AI across service workflows, up from 32%."](/guides/ads/1597821666.jpg) The figures are Freshworks' own report numbers. The library shows format, run dates, an impression band and targeting. It shows no clicks, no spend and no results. An outsider cannot say which container carried the idea. Only the advertiser can, and only if its account holds results for the same concept in each container. Public benchmarks do not close this gap. They compare formats across advertisers. Our reading: advertisers choose the format to suit the offer. A free guide goes into a document. A brand film goes into a video. So a format table compares offers as much as formats. Two examples: - A 2025 benchmark from an ad platform vendor pools advertisers and weights them by spend. In its table, lead generation campaigns with document ads have the lowest average cost per lead of four formats, and conversation ads have the highest. The page gives no sample size for each format. - Kiin, a LinkedIn Ads agency, compared formats inside the same accounts in [a report of 8 October 2026](https://kiin.co/research/linkedin-video-ads-benchmarks). For the median advertiser that ran both on the same objective, video cost 62% more per lead than an image on lead generation (95% interval from 12% to 210%). The report does not say whether the two ads carried the same creative and offer. This guide prints no format league table. To check your own account, list the ads of the last 12 months and write each ad's claim in one sentence. Mark every claim that ran in two or more containers. If no claim did, your account holds format habits and no format evidence yet. The last section shows how to build the evidence. ## Which LinkedIn formats exist, and which objective does each support? LinkedIn decides which formats you can pick by the objective you choose. The table follows [LinkedIn's page on ad formats for objective-based advertising](https://www.linkedin.com/help/lms/answer/a422880), which we read on 23 October 2026. The job applicants objective is left out. | Format | Brand awareness | Website visits | Engagement | Video views | Lead generation | Website conversions | |---|---|---|---|---|---|---| | Single image | Yes | Yes | Yes | No | Yes | Yes | | Carousel | Yes | Yes | Yes | No | Yes | Yes | | Document | Yes | Yes | Yes | No | Yes | Yes | | Video | Yes | Yes | Yes | Yes | Yes | Yes | | Thought leader | Yes | No | Yes | Video posts only | No | No | | Message | No | Yes | No | No | Yes | Yes | | Conversation | Yes | Yes | Yes | No | Yes | Yes | The table leaves out event, article and newsletter, text, spotlight and follower ads, so check Campaign Manager for those. LinkedIn's pages differ in places. Its [document ads page](https://business.linkedin.com/marketing-solutions/native-advertising/document-ads) names three objectives (brand awareness, engagement and lead generation), and the help page above adds website visits and website conversions. The table follows the help page. The video posts note for thought leader ads comes from LinkedIn's [thought leader ads page](https://www.linkedin.com/help/lms/answer/a1399568). Four facts matter most for a B2B team: - A thought leader ad has no lead generation objective. Image and video posts carry no button, so the link goes in the intro text. The post must be public, and the member must give permission. - A document cannot become a thought leader ad. The same page says documents, polls, celebrations, multi-image posts and reposts are not eligible. - A lead gen form attaches to single image, carousel, video, document, message and conversation ads. A carousel sends all its cards to [one form](https://www.linkedin.com/help/lms/answer/a427022), and [a carousel cannot hold a video](https://www.linkedin.com/help/lms/answer/a421188). - A document ad on the lead generation objective takes PDF (portable document format) files only. It gets one button, which LinkedIn labels "Unlock full document" in its [setup page](https://www.linkedin.com/help/lms/answer/a733950). The [specs page](https://www.linkedin.com/help/lms/answer/a493903) states the PDF rule. These are the specs we read for the five containers. Where a help page gives a higher maximum for a text field, the table shows the length that avoids truncation. | Container | File | Count and size | Text | |---|---|---|---| | Single image | JPG, PNG or GIF. 1.91:1 (1200 by 628 pixels), 1:1 (1200 by 1200) or 4:5 vertical (720 by 900). Vertical serves on mobile only | 5 megabytes (MB) | Headline 70 characters. Intro text 150 | | Carousel | JPG, PNG or still GIF cards at 1:1, 1080 by 1080 recommended | 2 to 10 cards, 10 MB each. No video | Card headline 45. Intro text 150 (255 maximum) | | Document | PDF, DOC, DOCX, PPT or PPTX. Lead generation takes PDF only | 100 MB and 300 pages. LinkedIn recommends under 10 pages | Headline 70. Intro text 150 | | Video | MP4 at 4:5, 9:16, 16:9 or 1:1. Captions as SRT files | 3 seconds to 30 minutes, 500 MB. LinkedIn recommends 15 to 30 seconds | Headline 70. Intro text 150 | | Thought leader | A member's public text, single image, video, article or newsletter post | Follows the single image or video specs | Link in the intro text | The sources are LinkedIn's spec pages for [single image](https://business.linkedin.com/advertise/ads/sponsored-content/single-image-ads-specs), [carousel](https://business.linkedin.com/advertise/ads/sponsored-content/carousel-ads/specs), [video](https://business.linkedin.com/advertise/ads/sponsored-content/video-ads/specs) and [document](https://business.linkedin.com/advertise/ads/sponsored-content/document-ads/specs) ads, and the help pages for [video](https://www.linkedin.com/help/lms/answer/85306) and [single image](https://www.linkedin.com/help/lms/answer/a426534) specs. ## Which kind of argument fits which container? Pick the container by what the argument needs from the reader. The table is our reading. It rests on LinkedIn's own use-case lists and on real ads. We found no public test that holds the concept fixed across containers, so treat the table as a first hypothesis and test it. | The argument needs | Container | What the buyer pays | A real ad | |---|---|---|---| | One claim and one proof that the buyer takes in at a glance: a customer quote, a number, an analyst badge | Single image | One glance, no commitment | [monday.com](/guides/ads/1578354254.jpg) | | Reasons or steps in a fixed order, where each step makes the next worth a swipe | Carousel | One swipe per card | [Adobe for Business](/guides/ads/1605684644.jpg) | | Reference material the buyer scans, compares or keeps: a framework, a benchmark, a checklist, a template | Document | Minutes of reading at the buyer's own pace, and a work email if you gate it | [monday.com](/guides/ads/1588880964.jpg) | | A person speaking or the product in motion, where trust comes from seeing and hearing | Video | Seconds, at the ad's pace | [Personio](/guides/ads/1585022103.jpg) | | A view that needs an owner, where the poster's name is part of the proof | Thought leader post | Attention, and a judgement about the poster | [Adobe](/guides/ads/1584237893.jpg) | | An invitation written to one person by name (outside the five) | Message ad | A reply or a registration | [GitLab](/guides/ads/1600409724.jpg) | The six real ads, coded by us: - Single image, monday.com: „Das Enterprise-Paket von monday.com erfüllt alle unsere Datenschutz- und Compliance-Vorschriften.“ (Our translation: the monday.com enterprise plan meets all our data protection and compliance rules.) A named customer answers one risk, data protection, with one proof. The reader needs one glance. - Carousel, Adobe for Business: the card headlines we can read are "Teams need AI-driven systems.", "Repurposing drains time." and "Three stages to scale." They run in order: the need, the cost, the plan. - Document, monday.com: a seven-page guide to resource planning. The next section takes it apart. - Video, Personio: the copy opens with a customer quote, „Echte Digitalisierung.“ ("real digitalisation"). The copy says the management board, the HR lead and support tell the story in the video. The proof is people. - Thought leader post, Adobe: a managing director asks „Wer ist in eurem Unternehmen eigentlich für KI verantwortlich?“ ("Who in your company is actually responsible for AI?"). The poster's role is part of the proof. The library lists it as an article ad with the poster's name on top. - Message ad, GitLab: "Would you join us?" The sender invites one person to a small event. The ads show what each container carries, not what worked. To read competitor ads the same way, see the [LinkedIn ad library guide](/guides/linkedin-ad-library). Our reading: two containers let the reader set the pace. A carousel and a document wait for the reader to swipe or scroll. They suit arguments that the reader scans, compares or returns to. A video sets its own pace, so it suits arguments that depend on timing, a voice or a product in use. A single image has no pace. It works when the claim survives one glance. LinkedIn's [ads guide](https://business.linkedin.com/advertise/ads/ads-guide) draws the same lines in its own words. Single image ads give "a clear message, a strong visual, and a single call to action". Carousels "walk through processes, or share a sequence of insights". Document ads "preview eBooks, guides, or whitepapers". Video ads "explain complex ideas, or show products in action". Thought leader ads "humanize your message". Some arguments need two containers. Split them. Use one container to reach the buyer and a second to carry the reference. LinkedIn's getting started guide for document ads lists retargeting members who interacted with a document ad. LinkedIn's [lead gen forms page](https://www.linkedin.com/help/lms/answer/a423447) says you can build a matched audience from members who engaged with a form. Name the angle of the argument first. The [angles guide](/guides/b2b-ad-angles) gives the words, and the [hooks guide](/guides/b2b-ad-hooks) covers the first line of each container. The [thought leader ads guide](/guides/linkedin-thought-leader-ads) covers who should post. ## How do you build a document ad that people read to the end? Build it as a short argument. Put the topic and the reader on the cover, give each page one idea, and put the first useful answer inside the pages that show before any form. The sources are LinkedIn's [getting started guide for document ads](https://business.linkedin.com/content/dam/me/business/en-us/marketing-solutions/resources/pdfs/linkedin-document-ads-getting-started-guide-v2.pdf) (a PDF whose file date is 14 April 2025), LinkedIn's help pages, and two real documents. ### What LinkedIn says about the file LinkedIn's [document ads page](https://business.linkedin.com/marketing-solutions/native-advertising/document-ads) gives no performance number. The getting started guide gives two. One is the advice to stay under 10 pages. The other is the line that gating a document with a lead gen form can drive "up to 4x more leads". The 4x line names no study and no comparison group. Read it as a sales line. The guide and the help pages give these file rules: - Use one PDF with flat layers. All pages need one page size. - Video and animation show as still images. - Links do not work in the feed viewer. They work after the reader downloads the file. - Use text of at least 8 point. LinkedIn says readers open documents mostly on a phone or a tablet, so test your pages on a phone. Our reading: 8 point is the floor, so aim larger. - A one-page file cannot sit behind a form. Add a cover page. - A saved ad cannot take a new document. Finish the document before you save the ad. Campaign Manager reports how much of the document readers saw (25, 50 and 75 percent), completions and download clicks. The displays figure counts only reading in the feed, not after a download. The same guide lists page views and view time under limits. You cannot see which page readers leave on. Test versions of the whole document, not single pages. ### Where to put the gate On the lead generation, website visits and website conversions objectives, a slider sets how many pages the reader can preview ([setup page](https://www.linkedin.com/help/lms/answer/a733950)). LinkedIn's guide says to keep 3 to 4 enticing preview pages, with helpful stats or a table of contents. Some vendor guides say to show 2 to 3. The two differ by one page. Real advertisers do not agree either. The library viewer for each document ad has a header such as "Preview 3 of 22 pages". Our reading: that number is the slider setting, because the viewer lets us page through only that many pages. The library does not show the objective, so a document with every page open may carry no gate. On 23 October 2026 the six documents in our first library pull read like this: | Advertiser | Pages | Pages open in the library | |---|---|---| | [Freshworks](/guides/ads/1597821666.jpg) | 74 | 1 | | [Salesforce](/guides/ads/1630858566.jpg) | 22 | 3 | | [Atlassian](/guides/ads/1644871936.jpg) | 20 | 5 | | [Snowflake](/guides/ads/1633878686.jpg) | 43 | 5 | | [monday.com](/guides/ads/1588880964.jpg) | 7 | 7 | | [Celonis](/guides/ads/1605518554.jpg) | 7 | 7 | The four long documents open 1, 3, 5 and 5 pages. Freshworks opens the cover alone, so its cover carries the whole promise. Six ads are a sample of what we pulled, not a measure of the market. Our reading: the preview must already hold one answer the reader can use, because the reader pays a work email for the rest. Do not gate weak, generic content for cold audiences: a gate on weak content asks for a work email and gives little back. The [lead quality guide](/guides/low-quality-leads-linkedin-meta-ads) covers what a weak gate costs. Test 3 pages against 4 with the slider as the only change. ### What two real documents do We paged through two document ads in the library on 23 October 2026. The viewer header read "Preview 7 of 7 pages" for both, so every page was open to us. The library does not show the objective. The [monday.com document](/guides/ads/1588880964.jpg) has seven vertical pages and is in German. The translations are ours. | Page | What the page does | Headline on the page | |---|---|---| | 1 | Names the topic and reframes the blame. Lists three daily pains under "Kennst du das?" (do you know this?) | „Dein Team ist nicht zu langsam. Es ist überplant.“ (the ad blames the plan and clears the team) | | 2 | Shows what the reader gets, as four numbered items, plus a skip-ahead sign-up line | „Das erwartet dich“ (what to expect) | | 3 | Gives the first framework: three levels, with a short set of questions for each | „Ressourcenplanung findet auf drei Ebenen statt“ (resource planning happens on three levels) | | 4 | Gives a 2 by 2 of four resource types, with one closing line | „Vier Ressourcentypen, die zusammenhängen“ (four resource types that depend on each other) | | 5 | Lists the first four of seven steps and says the other three need "mehr als eine Tabelle" (more than a spreadsheet) | „Die ersten vier von sieben Schritten“ (the first four of seven steps) | | 6 | Shows the product with sample data, labelled "Beispieldaten" | „Engpässe sehen, bevor sie bremsen“ (see bottlenecks before they slow you down) | | 7 | Makes one ask, and prints the web address as plain text | „Ressourcen planen. Ohne Überlastung.“ (plan resources, without overload) | Our reading of the structure: problem, map of the answer, first useful framework, body, proof, ask. The document gives away four of its seven steps and sells the tool for the rest. The address is plain text because links do not work in the viewer. The [Celonis document](/guides/ads/1605518554.jpg) has seven square pages and puts one idea on each. It runs as a story. The cover gives the promise („12 Monate schneller am Markt“, "12 months faster to market"). Then come the problem as one number (about 12 years to bring a drug to market), the cost of delay, the solution and three claims about it. A quote from a customer executive follows, and a photo carries the next step. It reads like a carousel inside a document. So a document is not always text-heavy. The two real files have two shapes: a reference (monday.com) and a story (Celonis). Choose the shape by the argument. Only these two files follow LinkedIn's advice of under 10 pages. The other four in our pull have 20 to 74 pages. ### The document ad template ```template DOCUMENT AD TEMPLATE (7 pages, gate after page 3) Fill every [bracket]. Delete this line. FILE RULES - Export one PDF. A lead generation ad takes PDF only. - Name the file as the headline, 70 characters or fewer. LinkedIn fills the headline from the file name. - Use one page size for all pages, square or vertical. Flatten layers. Use no video and no animation. - Links do not work in the feed viewer. Print the address as plain text. - Test every page on a phone before you upload. - A one-page file cannot be gated. Add a cover page. - You cannot change the document after you save the ad. PAGE 1, COVER (shows in the feed, so it must stop the scroll) Job: say who the document is for and what they get. Write: a label "[document type] for [reader role]" and one headline of 12 words or fewer. The headline names [the reader's problem] or [the result]. Check: the headline makes sense without the intro text above it. PAGE 2, WHAT YOU GET (preview) Job: let the reader judge the price of attention before they pay it. Write: 3 or 4 numbered items, each as [verb] + [thing]. Add "[n] pages" if it is true. PAGE 3, THE FIRST USEFUL ANSWER (preview, the last page before the gate) Job: give one finding, rule or framework that is useful on its own. Write: a headline with [the answer in one sentence]. Then [3 short items or one small table]. Last line: [what the next pages add]. PAGES 4 AND 5, THE REFERENCE (behind the gate) Job: the part the reader keeps. Write: one idea per page, as [steps, a checklist, a table or numbers]. Every number names its source and date. PAGE 6, PROOF Job: show the idea at work. Write one of these: [a customer quote with name, role, company and written permission], [a product screen with sample data labelled as sample] or [a result with source and date]. Never invent a quote. PAGE 7, NEXT STEP Job: one ask. Write: a headline with [the outcome]. One line for the ask: [ask]. The address as plain text: [short web address]. One line on what happens next: [what the reader gets, and when]. AD FIELDS Headline (70 characters or fewer): [file name] Intro text (150 characters or fewer): [who it is for, what is inside, and that more pages follow] Preview pages: [3]. Test [4] against it. Objective: [lead generation with a lead gen form] or [website visits, with a destination URL] ``` ## How does one concept look in all five containers? One concept can run in all five containers when you hold the claim, the proof, the offer and the ask fixed and change only the form. The example below uses Acme Approvals, an invented invoice approval tool for finance teams. The product, the reader and the claim are constructed for this example, and every proof field is a placeholder. The [brief template guide](/guides/b2b-creative-brief-template) shows where a real concept card comes from. Four of the five containers can run on the lead generation objective. The thought leader post cannot. It runs on brand awareness or engagement, and its ask becomes a click on a link in the text that leads to the same checklist. Start with a concept card. Fill it before you brief any container. ```template CONCEPT CARD Name: [short name, used in the ad names] Reader: [role and company type] Risk the reader carries: [what goes wrong for them if they pick wrong or do nothing] Claim: [one sentence the reader can repeat to a colleague] Proof: [named customer, result, source, date] or [none yet] Offer: [what the reader gets] Ask: [what the reader does, and what it costs them: a click or a work email] Objective and goal metric: [the objective you buy] and [the one result you judge] Poster (thought leader post only): [name and role of a real person at the company] MUST STAY EQUAL IN EVERY CONTAINER Claim, proof, offer, audience, start date, end date Ask and objective too, except for the thought leader post, which cannot run on lead generation MAY DIFFER Length, structure, picture, voice ``` Here is the card for Acme Approvals. It is constructed. The five versions below are illustrative: we wrote them by hand to show the shape. Numbers in parentheses after a field name are the character limits we read. ```text CONCEPT CARD (constructed example) Name: audit-question Reader: head of accounts payable at a mid-sized company Risk the reader carries: pays an invoice that nobody can show was approved, then has no answer when an auditor asks Claim: every invoice approval carries a name and a date Proof: [named customer, role, result in their words, source, date] Offer: the approval audit checklist, with the questions an auditor asks about invoice approvals Ask: leave a work email in a lead gen form Objective and goal metric: lead generation, cost per lead Poster (thought leader post only): [name and role of a real person at the company] ``` ### Single image ad ```text Image text: Who approved this invoice? / Nobody remembers. Proof line on the image: [customer name, result, source, date] Headline (70): The invoice approval audit checklist Intro text (150): An auditor asks who approved an invoice. Your inbox does not know. Get the questions auditors ask, with room for your answers. Button: Download Form promise: the checklist arrives by email at once ``` ### Carousel ad ```text Intro text (150): Five cards on a question an auditor asks about invoices. The checklist is on the last card. Card 1 (45): Who approved this invoice? Job: the promise Card 2 (45): Email threads hide the answer. Job: a reason to swipe, because it shows in part next to card 1 Card 3 (45): An auditor needs a name and a date. Card 4 (45): [Customer]: [result], [source, date] Card 5 (45): Get the approval audit checklist. Job: the offer Button: Download. Every card opens the same lead gen form. ``` LinkedIn's [carousel tips](https://business.linkedin.com/advertise/ads/sponsored-content/carousel-ads/tips) say the first card shows in full and the second shows in part. They also say to start with 3 to 5 cards and, when the carousel tells a story, to put the main value or the call to action in the last card. That page reads as old, so check it against your own export. It says the downloadable report shows clicks and impressions for each card. ### Document ad ```text Headline (70): Invoice approval audit checklist Intro text (150): The questions an auditor asks about invoice approvals, with room for your answers. Read the first pages, then get the rest. Page 1: Label "Checklist for heads of accounts payable". Headline: Who approved this invoice? Page 2: What you get. 1 The [n] questions an auditor asks about one invoice. 2 A table for your answers. 3 A check you can run on one team. 4 What to fix first. Page 3: Three facts make an approval you can show: a name, a date and the invoice it covers. [Gate after this page.] Page 4: The questions, as a table: question, and where the answer should live. Page 5: The three places an approval gets lost: an inbox thread, a chat message, a spreadsheet. Page 6: [Customer quote with name, role, company and permission] Page 7: Run the check on one team. Keep your finance system. [short web address as plain text] ``` ### Short video ad The video runs 20 seconds at 4:5 with burned-in subtitles. | Seconds | On screen | Voice-over | On-screen text | |---|---|---|---| | 0 to 2 | A search box in an inbox: "approved invoice 4471". The result list is empty. | None | Who approved this invoice? | | 2 to 7 | The same search in a chat app, then a spreadsheet. | "The auditor asks. You search three places." | Three places. No answer. | | 7 to 13 | An Acme Approvals screen: the invoice with the line "Approved by [name], [date]". | "In Acme Approvals, every approval carries a name and a date." | A name and a date on every invoice. | | 13 to 17 | A proof card. | [Customer] says: [quote]. | [Customer]: [result], [source, date] | | 17 to 20 | End card with the offer. | "Get the approval audit checklist." | Get the checklist | LinkedIn loops a video under 30 seconds until it has played for at least 30 seconds, per its [video specs help page](https://www.linkedin.com/help/lms/answer/85306). A 20-second video plays twice. So the end card sits next to the first frame. Make the last line and the first line work together. ### Thought leader post ```text Poster: [name and role of a real person at the company]. Objective: brand awareness or engagement. No button and no lead gen form. Who approved this invoice? [One real moment from the poster's own work, with the date.] The answer was in [where it was]. Now I ask one question about every approval: does it carry a name and a date? The checklist we use is here: [link to a page that lets the reader download it] ``` Only the poster can supply the real moment. A model cannot, and nobody should invent it. ### The five-container prompt This prompt turns a filled concept card into the five versions. It works in Claude or ChatGPT. Paste your own card at the end. ```prompt You are a B2B paid social copywriter for LinkedIn. Turn the concept card at the end into one version of the same concept in five containers: a single image ad, a carousel ad, a document ad, a short video ad and a thought leader post. Rules 1. The claim, the proof, the offer and the ask stay the same in all five. Only the form changes. One exception: a thought leader post cannot use a lead gen form, so its ask is a click on a link to the same offer. 2. Use only the proof I give. If the proof says "none yet", write [PROOF NEEDED] where a proof would go. Never invent a customer, a quote, a number or a source. 3. Write plain, specific sentences. No dashes. No filler adjectives. 4. Respect these limits. Count the characters of every limited field and write the count after it. Shorten any field that is over. Single image: headline 70 characters, intro text 150. Also give the text that goes on the image. Carousel: 3 to 5 cards, card headline 45, intro text 150. Card 1 gives the promise. Card 2 gives a reason to swipe. The last card holds the offer. Document: headline 70, intro text 150, and 7 pages. For each page give a one-line job and the page text. Pages 1 to 3 show before the gate, so page 3 must give one useful answer. Video: 15 to 30 seconds. Give a table with columns for seconds, what is on screen, voice-over and on-screen text. The claim must read in the first 2 seconds with the sound off. Plan burned-in subtitles. The last frame must work next to the first frame, because the video loops. Thought leader post: write it in the first person for the poster named in the card. If the card names no poster, write [POSTER: name and role] at the top. It has no button and no lead gen form. Put the link in the text. Where only the poster can know a fact, such as a real moment from their work, write [POSTER: add a real moment]. 5. After the five versions, write a table with one row per container and columns for claim, proof, offer and ask. The cells must match. If a container cannot carry the claim, the proof, the offer or the ask as written, say so. Then list what each container forced you to change. CONCEPT CARD [paste your concept card here] ``` Check the output against rule 1 before you brief anything. If the claim changed in one container, send that container back. Models miscount characters, so paste each field into the ad preview in Campaign Manager and check where the text cuts off. ### What is the hook window of a video? Sources disagree on how soon the argument must land. We show the disagreement and do not pick a winner. | Source | Window | What it is | |---|---|---| | LinkedIn, [video metrics page](https://www.linkedin.com/help/lms/answer/a426666) | 2 seconds | The paid view definition: two continuous seconds of playback with the video at least 50% on screen | | Adobe Express, [blog post of 2025](https://www.adobe.com/express/learn/blog/video-ads-linkedin) | 3 seconds | The vendor's advice | | LinkedIn, [video ad tips](https://business.linkedin.com/advertise/ads/sponsored-content/video-ads/tips), for a video of about 30 seconds | 5 seconds | "Establish your point in the first 5 seconds" | | LinkedIn, the same tips page, in general | 10 seconds | "Show what you want your audience to see in the first 10 seconds" | The same tips page says short videos of 7 to 15 seconds see "up to a 300% lift in completion rates" (LinkedIn data, 2025, method not stated). It also says longer videos drove as many clicks and views. It says a large share of members watch with the sound off, and it suggests burned-in subtitles. Our reading: LinkedIn counts a view at two seconds, so write the claim to read in the first two seconds with the sound off. Then land the payoff by second five, the window LinkedIn names for a video of about 30 seconds. ## How do you test formats without testing the concept by accident? Keep the claim, proof, offer, ask, audience and objective equal, change the format, and run the pair in LinkedIn's A/B test. LinkedIn says the test splits the audience into two groups, and each group sees one ad set. A few members can land in both groups, and LinkedIn calls that effect generally insignificant. Two ordinary ad sets with the same targeting both aim at the whole audience, so the same members can see both ads. ### Use LinkedIn's A/B test with the ad variable LinkedIn's [ad variable test](https://www.linkedin.com/help/lms/answer/a1376579) can compare two formats or two versions of one format. Both ad sets share every setting except the variable. The [A/B testing pages](https://www.linkedin.com/help/lms/answer/a529753) and [best practices](https://www.linkedin.com/help/lms/answer/a525922) give these rules: - A test compares two ad sets. A question about five containers becomes a ladder of pair tests. - A test runs at least 14 days and at most 90. LinkedIn recommends 21 days. - The minimum budget is $700 lifetime or $20 a day. The pages name no currency and do not say whether the figure is for each ad set, so read the budget screen in your account. - LinkedIn recommends at least 300 members per ad set, and one ad per ad set. - If you change or remove an ad in the winning ad set, the result is no longer statistically valid. - Results appear as soon as two days after the start. LinkedIn still asks for two weeks. When the result is directional or inconclusive, LinkedIn offers to extend the test by two weeks ([results page](https://www.linkedin.com/help/lms/answer/a1449372)). The P-value can keep moving for seven days after the end date ([result categories](https://www.linkedin.com/help/lms/answer/a1454650)), so wait those days before you call the result final. - LinkedIn calls a P-value of 10 percent acceptable. When two containers do not differ, about one test in ten still shows a winner at that level. If no container is better, a ladder of four pair tests then crowns at least one false winner about one time in three (1 minus 0.9 to the power of 4 is 0.34). Treat a directional winner as a lead, and repeat the test before you rebuild a plan on it. - In the European Economic Area and Switzerland, cost per paid conversion and cost per qualified lead count only members who gave consent. Start the ladder with the two containers that fit your argument best in the chooser table. The winner meets the next container in a later test. The thought leader post joins only a test on brand awareness or engagement, because it has no lead generation objective. ### Keep five things equal 1. The concept: claim, proof, offer and ask. Read both ads aloud and check. 2. The destination: the same lead gen form or the same page. 3. The audience, objective, bid strategy and placement. 4. The craft: build both formats to one standard, with one owner and one review. A produced video against a quick image tests the budget. 5. The dates: both ad sets start and end together. Write the one result you will judge before launch. Compare formats on the result the objective buys, because a click is a different act in each format. LinkedIn's getting started guide for document ads marks the first click on the page viewer, and on full screen, as chargeable on engagement objectives. A reader who pages through your document has clicked. A reader who clicks a single image ad usually leaves for your page or form. ### Mistakes that fool a format test - Two offers. A free guide in the document and a demo request in the video test the offer. - LinkedIn's AI variations and flexible ads, which create several versions at once. The test then compares more than the format. - Edits during the test. A saved document ad cannot change, and an edited ad breaks the test. - A stop on the first good numbers. Fix the end date in advance. The [pause guide](/guides/when-to-pause-an-ad) shows how to choose the numbers. ### The format test sheet ```template FORMAT TEST SHEET Test name: [concept name], [format A] against [format B] Question: Which format carries the concept better for [audience]? Concept card: [paste it or link it]. Claim, proof, offer and ask are identical in both ad sets. SAME IN BOTH AD SETS Objective: [one objective that both formats support] Result I will judge: [the one result the objective buys, written before launch] Audience: [the same saved audience, at least 300 members] Budget: [the test budget, split evenly between the two ad sets]. LinkedIn lists a minimum of $700 lifetime or $20 a day for a test. Check the budget screen. Bid strategy: [same] Placement: [same, for example the LinkedIn feed only] Dates: start [date], end [date]. At least 14 days. LinkedIn recommends 21. Ads per ad set: [1] Destination: [the same lead gen form or page] DIFFERENT Format A: [format]. Ad name: [concept]_[formatA]_v1 Format B: [format]. Ad name: [concept]_[formatB]_v1 Craft: [who built each, to which standard] CHECK BEFORE LAUNCH [ ] I read both ads aloud and the claim, proof, offer and ask match. [ ] Both ad sets use the same destination. [ ] The result I will judge can be read for both formats. [ ] Nobody edits an ad until the end date. AFTER THE END DATE Result: [clear winner, directional, or no clear winner] Decision: [duplicate the winner, extend two weeks, or drop the format for this concept] What we learned about the concept and about the format: [one sentence each] Next test: [one change] ``` Name every ad with the concept and the container, for example audit-question_document_v1. Any tool can then group ads by concept. Nimblet types each synced LinkedIn ad as single image, carousel, document, video, thought leader and so on, and it keeps the ad name. A naming convention or a contains rule can turn the concept part of the name into a tag. Through the [Nimblet connector in Claude or ChatGPT](/mcp) you can then group results by that tag and by ad type, and rank on the goal each campaign was bought for. To see your own ads lined up this way, read about [LinkedIn ads creative analytics](/linkedin-ads-creative-analytics). When the result is ready, the [sharing guide](/guides/share-creative-learnings) shows how to report it. ## Nimblet runs this check Nimblet types the LinkedIn formats you run, such as single image, carousel, document, video and thought leader ads. Ads with recent spend also get an AI read of hook, offer and proof, so you can line up ads with the same core message in different formats. Nimblet ranks them on the goal each campaign was bought for, and it only reads your ad accounts and never changes a campaign. ## Frequently asked questions ### Which LinkedIn ad format is best for B2B? No format is best across arguments. The argument decides. One claim with one proof suits a single image, and steps in order suit a carousel. A reference suits a document, and a person or a product in motion suits a video. Start with the chooser table, then test two formats with the concept held fixed. ### Should a document ad be gated? Gate it when the pages before the gate already give a useful answer and the rest is reference the reader keeps. Leave it open when you want reach. LinkedIn says gating can drive up to 4x more leads and names no study. A version gated by a lead gen form runs on the lead generation objective and an open version on another, so the two buy different results. LinkedIn tells you to pick objectives that make sense for the test metric. To test where the gate sits, compare 3 and 4 preview pages on the same objective. ### How many pages should a document ad have? LinkedIn recommends under 10 pages and allows 300. The six document ads in our library pull have 7, 7, 20, 22, 43 and 74 pages. Four of the six go past LinkedIn's advice. Give each page one job, and cut the page that has none. ### How long should a LinkedIn video ad be? LinkedIn allows 3 seconds to 30 minutes and recommends 15 to 30 seconds so the ad can run in every placement. LinkedIn's tips say 7 to 15 second videos see up to a 300% lift in completion rates, and the tips give no method. Sources disagree on the hook window, from 2 to 10 seconds. Write the claim to read in the first 2 seconds with the sound off. ### How many cards should a carousel have? LinkedIn allows 2 to 10 cards. Its tips say to start with 3 to 5 and to test longer carousels from there. When the carousel tells a story, put the main value or the call to action on the last card. ### Can I use a lead gen form with a thought leader ad? No. LinkedIn lists brand awareness and engagement for a thought leader ad, and video views for a video post. It has no lead generation objective. Use a document, carousel or single image ad when the form is the goal. ### Can a carousel hold a video? No. LinkedIn's help page says videos cannot be added to carousel image ads. A carousel takes still images. ### Do links work inside a document ad? Not in the feed viewer. They work after the reader downloads the file. Print the address as plain text, as the monday.com document does. The Celonis document names its site in plain text on the last page. ### Which formats can LinkedIn's A/B test compare? The ad variable compares two ad formats, or two versions of one format, in a pair of ad sets. A test needs at least 14 days. Test five containers as a ladder of pairs. ### Does Nimblet read every page of a document ad? No. Nimblet reads the first page of a document ad as an image. It also syncs document completions and download clicks. It does not read pages 2 and up.