The underused B2B surface
Scripts written for the transcript, not the camera
A video is invisible to a retrieval system until it becomes text. The transcript is the asset, it sits on a domain Google owns, and it is indexed prose whether four people watched or four thousand did. Which means the script decides almost everything and the production decides almost nothing.
Every plan queries all 6. See how each engine picks who it names
What we measured
as many B2B buyers now name AI as a top information source as did before (Forrester, 2026), and the Google-side surfaces that answer them draw on an index that includes YouTube.
Engines queried
ChatGPT, Claude, Perplexity, Gemini, Google AI Mode, Microsoft Copilot. Every plan, including the free one.
Answers kept
Every answer is stored word for word, so any number on your report can be checked against the text it came from.
The gap this closes
- Video is treated as a brand exercise, judged on views, and stopped when views disappoint.
- A video is invisible to retrieval until it becomes text.
- A watch-time script delays the answer to hold attention, which is the opposite of what indexing wants.
- Almost nobody writes the retrievable kind, so the competition is close to nothing.
What you end up with
- A script built for a clean transcript that still works as a video.
- The answer stated early, in indexed prose, on a domain Google owns.
- Description and chapter plan carrying the same answer as text.
- An asset that works whether four people watched or four thousand.
How it works
Three steps, and you only do the first one.
The episode targets a question you are losing
Video briefs come from the same ranked list as the written work: buying questions where an engine named somebody else. Each brief carries a hook and a beat-by-beat outline aimed at that phrasing, so the episode has a target before a word of script exists.
The script is written to transcribe cleanly
Spoken filler dilutes a transcript, so the script cuts it. The question gets stated aloud, the answer lands next, and each beat is a self-contained passage that survives being pulled out of the video with nothing around it.
Description and chapters carry the same answer
The description is written as a real summary containing the question and the answer, because it is indexed text in its own right. Chapter titles are phrased the way buyers phrase sub-questions, which gives the transcript structure that headings would otherwise provide.
What you get on each plan
You get a script built for a clean transcript, plus the description and chapter plan that carry the same answer as indexed text. This is what lands, per plan, with the monthly volume read live from the plan configuration.
Track
Free- No scripts, and the harder half anyway
- The buying questions where a competitor was named
- The full answers each engine gave
- The topic list a paid plan would draft
- Tells you which episode to make next
Fix
$99/mo- Video briefs alongside the written ones, on the report
- A real hook and a beat-by-beat outline
- Aimed at a question you are currently losing
- Finished scripts are a Dominate deliverable, briefs are the output here
- Worth knowing before you buy, not after
Dominate
$499/mo- Briefs become finished scripts
- Spoken opening written out, beats in order
- Description written as a summary, not a link dump
- Chapter titles phrased how buyers ask
- Video and article answer the same question across two domains
Rule
$1999/mo- Portfolio level, where video suits one or two brands
- A person decides which brands earn camera time
- Cadence set to what those brands can sustain
Which engine actually leans on this
Your row is Video. The rest of the grid is on the engines page.
| Channel | ChatGPT | Claude | Perplexity | Gemini | Google AI Mode | Microsoft Copilot |
|---|---|---|---|---|---|---|
| Your own site Schema, answer pages, llms.txt, crawlability | Heavy Vendor | Heavy Vendor | Heavy Measured | Heavy Our read | Heavy Vendor | Heavy Our read |
| Reviews and directories G2, Clutch, Capterra, Trustpilot, Google Business | Moderate Our read | Light Measured | Heavy Our read | Moderate Our read | Moderate Measured | Moderate Measured |
| Reddit and forums Unpaid opinion, where buyers go for the unvarnished view | Light Our read | Light Measured | Heavy Our read | Moderate Our read | Heavy Measured | Moderate Measured |
| Video YouTube. Only counts once there is a transcript | Moderate Our read | Light Our read | Moderate Our read | Heavy Our read | Heavy Measured | Light Measured |
| Newsletters LinkedIn, Substack, Beehiiv. Published under a named person | Moderate Our read | Light Our read | Moderate Our read | Heavy Our read | Heavy Measured | Light Our read |
| News and PR Earned coverage, wire pickup, journalist mentions | Heavy Vendor | Heavy Vendor | Heavy Our read | Moderate Our read | Moderate Our read | Moderate Our read |
Your own site. Your own site counts most for ChatGPT, Claude, Perplexity, Gemini, Google AI Mode, Microsoft Copilot.
Reviews and directories. Reviews and directories counts most for Perplexity. It is a light signal for Claude.
Reddit and forums. Reddit and forums counts most for Perplexity, Google AI Mode. It is a light signal for ChatGPT, Claude.
Video. Video counts most for Gemini, Google AI Mode. It is a light signal for Claude, Microsoft Copilot.
Newsletters. Newsletters counts most for Gemini, Google AI Mode. It is a light signal for Claude, Microsoft Copilot.
News and PR. News and PR counts most for ChatGPT, Claude, Perplexity.
Where this comes from
- OpenAI: publishers and developers FAQ States that a site must allow OAI-SearchBot to be eligible for ChatGPT search at all.
- OpenAI: introducing ChatGPT search The named publisher partnerships behind the news and PR row.
- Anthropic: web search tool documentation Citations are always enabled, and Claude only searches when a question needs current information.
- Google Search Central: AI features and your website AI Overviews and AI Mode report inside the ordinary Search Console Performance report.
- Search Console Help: impressions, position and clicks How AI Mode counts a click, and why a follow-up question is treated as a new query.
- Dan Petrovic: citation mining of OpenAI grounding metadata Reddit retrieved in 76 percent of OpenAI searches and selected 0.61 percent of the time.
- Our production scan database, read 12 August 2026 46 completed reports, 34 brands, 419 buying questions, 2,421 answered buying cells. Where the per-engine mention rates come from.
- Our 15-region managed services study 743 records, 566 named by an engine, and the per-engine corroboration rates.
- What our crawler finds on real sites 39 readable crawls: 38 with no llms.txt, 26 with no FAQ schema, 16 with no Organization schema.
- How we check an AI visibility report The method behind every first-party number on this page, including what it cannot show.
Engine behaviour moves month to month. This table is maintained, not published once.
The proof, including the parts that flatter nobody
Here is the honest position. Across the 46 completed reports behind this page, the best any engine managed was naming the brand in 7.4% of the buying questions we put to it, on firms that arrived already suspecting they were missing, so the base rate this work starts from is low and we can evidence that part.
What we cannot evidence is video attribution. We do not instrument YouTube, we do not know which passage an engine read, and we will not manufacture a number to fill the gap. The measurable claim is the question the episode targeted, re-asked on all six engines, scan after scan.
Our published research is where the method behind those figures sits, including the parts that did not go our way.
Read the MSP AI Visibility Report 2026 →A worked example: the first ninety seconds of a script
This is the opening of a script written for a firm losing the question of whether an outsourced provider can cover after-hours incidents for a manufacturer. The brand is illustrative. The structure is the structure every script ships with.
Spoken opening, written out
[0:00] Can an outsourced IT provider actually cover a plant at 2am?
Yes, and there are four things to check before you believe anyone
who says so. I will name all four, then show you what a real
after-hours ticket looks like from our side.
[0:18] One. Ask who picks up. Not which team. Which person, at 2am, on a
Sunday. If the answer is a rota you cannot see, it is a queue.
[0:41] Two. Ask for the contractual response window in writing, and ask
what happens when it is missed. A commitment with no consequence
is a marketing number.
[1:05] Three. Ask how many after-hours incidents they closed last quarter
without escalating to your staff. That single number tells you
whether coverage is real or nominal.What ships with it
- Chapter plan: Can a provider cover 2am / Who actually picks up / What the response window means / The escalation number to ask for / What an incident looks like end to end.
- Description written as a summary that states the question and the four checks in plain text, with one link to the matching page on your site.
- The same four checks appear on the written page and in the FAQ block, so the two surfaces agree word for word on the answer.
Read the opening again as text with the video removed. The question is stated, the answer is committed inside two sentences, and each numbered check stands alone as a passage. That is the property being engineered, and it costs nothing in watch time to have it.
How it actually works
The mechanism, for anyone who wants it. Open a card to read the detail.
What a transcript looks like to a retrieval system
Strip the video away and what remains is a wall of prose with no headings, no paragraphs and no markup. A retrieval system chunks that wall into passages and judges each passage on its own. So the unit of value is not the episode. It is the passage, and every passage has to make sense to somebody who never saw the other twenty minutes.
Read the detail
This is why rambling is expensive and polish is cheap. A tightly shot studio piece that circles its point for four minutes produces four minutes of unusable chunks. A phone recording that states a question and answers it in two sentences produces a chunk worth quoting. The script is doing all the work the layout would do on a web page.
It is also why the chapter plan earns its place. Chapter titles become visible structure on the platform and give a reader, human or otherwise, a map of which passage answers which sub-question. That is the closest a transcript gets to having headings.
Two surfaces answering one question
The reason a script comes from the same brief as an article is corroboration. A claim that appears once, on your own domain, has a single source behind it. The same claim on your site and in a transcript on a second high-authority domain is a claim two independent places agree on, and agreement is what a system checks before it puts a name in an answer.
Read the detail
Video is one of the cheapest independent platforms to add, because you are reusing an answer you already wrote instead of researching a new one.
The practical version is dull and effective. One buying question, one article, one episode, one chapter whose title matches the question, and a description that states the answer in plain text. Nothing here needs a content strategy. It needs the same answer to exist in two places a machine reads.
Where this helps, and where we will not oversell it
The engines this reaches most directly are the Google-side ones, because Google owns YouTube and the index that includes it. Our own measurement supports treating engines as separate problems rather than one: in our managed IT study, Perplexity resolved 101 of 257 of the firms it named to real websites while an engine answering from memory resolved 2 of 129. Different engines fail in different ways, and a second surface hedges you across them.
Read the detail
What video will not do is move an engine that answers from parametric memory. That kind of engine changes with corroboration accumulated over long periods, and one episode is not corroboration. Anybody promising you otherwise is guessing.
We also do not measure whether a published video earned a citation. Nothing in our product instruments a platform we do not host. What the scan can show is narrower and real: the question the episode targeted, re-asked on all six engines, before and after.
What it costs
Prices and monthly volumes below read from the live plan configuration, so this page cannot drift from what your account gets.
| Plan | Price | Questions | Scans |
|---|---|---|---|
| Track | Free | 5 | 1/mo |
| Fix | $99/mo | 10 | 4/mo |
| Dominate | $499/mo | 25 | 4/mo |
Rule is portfolio level and sales-assisted, which suits a group where one or two brands carry the video programme.
See where you stand first
Run a free scan and find out which of these gaps you actually have.
Get your free auditFrequently asked questions
Do AI engines really use YouTube content?
Through transcripts, yes. A transcript is indexed text on a domain Google owns, and the Google-side engines draw on that index. The video file itself is not the asset; the words in it are.
Do I need professional production?
No, and polish can work against you. Clarity produces a stronger transcript than a glossy video that circles its point, because a retrieval system reads passages and skips the ones that say nothing.
I hate being on camera. Does this still work?
Yes. The retrievable asset is the transcript, so a screen recording, a slide walkthrough or a plain voiceover carries the same words into the same index. The scripts are written for whichever format you will record.
Do my videos need to rank in YouTube search?
No. The citation path runs through the transcript and description as indexed text, which works whether or not the video ranks. Ranking brings human viewers, which is a separate and welcome benefit.
Which plan drafts the scripts?
Dominate. Paid scans below that produce video briefs with a hook and a beat-by-beat outline on the report, and the finished scripts sit in the higher tier. The plan table above reads the live configuration, so it stays accurate.
How would I know if a video moved anything?
By the question it targeted. We re-ask that question on all six engines every scan and show the before and after. We do not claim to attribute a citation to a video, because we do not instrument YouTube.
Sources and further reading
- Forrester B2B Buying Study 2026: 94% of B2B buyers use AI in the buying process; 2x now cite AI as a top information source.
- Lewis et al. 2020, Retrieval-Augmented Generation (arXiv:2005.11401): The foundational paper describing how models retrieve from an external, non-parametric memory to generate factual, sourced answers.
Related solutions