The YouTube algorithm is the set of systems that decide which videos each viewer sees: search ranks results by relevance, engagement and quality, while the home page, suggested videos and the Shorts feed are driven by what someone watches, for how long and whether they were satisfied. On 13 October 2026, across 8 Italian searches in our clients' sectors, the first video had a median of 7,805 views, while the most-watched video on the same page had 181,387: 23 times more.
That gap is the story.
Most guides on the YouTube algorithm are written for creators. Shopify, Sprout Social and vidIQ cover subscribers, the first 30 seconds and how often to post. Search Engine Land's piece is aimed at marketers, but it dates from February 2023.
And a company that sells LED walls or solar batteries?
It has a different problem. It does not need a million views. It needs its video to appear when an engineer searches "how does an LED wall work", or when a landowner searches "how does agrivoltaics work".
The three mistakes we see most often when a business asks about its channel:
- it judges videos like a creator would, on total views, and writes them off at 400;
- it chases tricks YouTube itself plays down, such as tags or daily uploads;
- it reads old figures as current ones: "70% of watch time comes from recommendations" was a 2018 statement, and "80 billion signals" comes from a 2021 blog post.
We work on Google search for companies that sell technical products. For Macropix, an Italian LED wall manufacturer, Google Ads has brought in 2,906 quote requests since 2018. For Valore Solare, which installs solar panels and storage, 1,506 enquiries.
Small markets, high-value customers. We wanted to see how YouTube behaves exactly there.
In this guide:
- the two systems behind the algorithm, explained from YouTube and Google Research sources;
- our test on 8 sector searches, with the full table;
- what YouTube says about the most repeated myths, from tags to upload frequency;
- five moves for a business channel that aims at search, not virality.
Let's go.
What the YouTube algorithm is (and why it is at least two systems)
The YouTube algorithm is not one formula: it is a set of machine learning models that pick, out of millions of videos, the few to show one person at one moment. The most cited technical description is Deep Neural Networks for YouTube Recommendations by Paul Covington, Jay Adams and Emre Sargin (Google, RecSys 2016).
The paper describes two neural networks in sequence:
- candidate generation, which pulls a few hundred videos from a catalogue of millions;
- ranking, which orders those hundreds and shows a few dozen.
The detail almost nobody quotes is the ranking objective. According to the authors it is "generally a simple function of expected watch time per impression". Not click-through rate.
Why not?
Because, they write, ranking by click-through rate "often promotes deceptive videos that the user does not complete". Clickbait was already a design problem ten years ago.
The first system to keep separate is search, which answers a typed query. The second is recommendation, which fills the home page, "Up next" and the Shorts feed when nobody has typed anything. That split runs through all of YouTube SEO: titles and descriptions move the first, much less the second.
What YouTube search looks at (three stated criteria, none of them for sale)
YouTube search ranks results on three elements, weighted differently depending on the search. YouTube Help says so on the page How YouTube Search works, checked on 13 October 2026:
- relevance: how well the title, tags, description and video content match the query;
- engagement: among other signals, the watch time of a video for that specific query;
- quality: signals that show which channels demonstrate expertise, authoritativeness and trustworthiness on a topic.
Two more lines from the same page matter to a business. If history is on, results may reflect past searches and views, so two people see different pages. And YouTube does not accept payment for better placement in organic search results. Ads exist, but they sit in separate slots.
View count is not listed as a criterion. Watch time for the query is, meaning how much the people who searched for that thing liked the video. How to write the fields that feed relevance is covered in YouTube titles, descriptions and tags.
Home, suggested and Shorts: what recommendation looks at (clicks, time, star surveys)
The recommendation system decides what to show without a query. YouTube's page on recommendations lists watch behaviour, likes, dislikes, subscriptions and feedback, plus channel reputation and quality, with input from external raters.
The most detailed document is On YouTube's recommendation system by Cristos Goodrow, VP of Engineering at YouTube, published on 15 September 2021. It says recommendations drive more viewing than channel subscriptions or search, without giving a percentage. And it explains the signals:
- clicks: a strong indication the video will be enjoyed;
- watch time: which videos you watch and for how long;
- surveys: viewers rate videos from one to five stars, and only four- and five-star videos count as "valued watchtime";
- shares, likes, dislikes: on average, people who share or like a video are more satisfied.
On length, YouTube's performance FAQ is specific: relative watch time matters more for short videos, absolute watch time for longer ones. A long technical video should not be judged on the percentage watched.
In Analytics, Shorts are a traffic source of their own, separate from browse and suggested. How to use them without turning the channel into an entertainment feed is in YouTube Shorts for business.
Our test on 8 sector searches (the top video was not the most watched in 6 cases out of 8)
On 13 October 2026 we pulled YouTube's own search results with DataForSEO for Italy, in Italian, with no history or account, for 8 searches in the sectors of clients whose case studies we publish. For each search we took the first 10 videos (playlists excluded; "ledwall prezzo" returned 9) and recorded views and upload date. The data is Italian; the pattern is what matters.
| YouTube search (Italian, with translation) | Sector | Views of video #1 | Most-watched video (position) | Videos under 1,000 views | Videos uploaded in the last 12 months |
|---|---|---|---|---|---|
| come funziona un ledwall (how does an LED wall work) | LED walls | 6,133 | 79,683 (2nd) | 2 | 3 |
| ledwall prezzo (LED wall price) | LED walls | 79,684 | It is #1 | 1 | 1 |
| batteria accumulo fotovoltaico come funziona (how does a solar battery work) | Solar | 6,735 | 202,752 (5th) | 1 | 3 |
| agrivoltaico come funziona (how does agrivoltaics work) | Agrivoltaics | 36,726 | It is #1 | 5 | 4 |
| affittare terreno per fotovoltaico (lease land for solar) | Agrivoltaics | 8,875 | 160,022 (2nd) | 0 | 5 |
| ricambi vespa px (Vespa PX spare parts) | Scooter parts | 12,609 | 215,246 (7th) | 0 | 2 |
| come scegliere offerta luce e gas (how to choose an energy tariff) | Energy | 3,784 | 217,224 (8th) | 2 | 5 |
| impianto fotovoltaico aziendale (commercial solar system) | Solar | 2,134 | 665,255 (6th) | 1 | 3 |
| Total | 79 videos | Median 7,805 | Median 181,387 | 12 of 79 | 26 of 79 |
Three things stand out.
Popularity does not set the order. In 6 searches out of 8 the first video is not the most watched on the page. For "commercial solar system" the first result had 2,134 views; sixth place went to a video with 665,255.
A video with few views can reach the first page. 12 of 79 videos had under 1,000 views. For "LED wall price" an Italian supplier, Getled Srl, was fourth with 34 views. For "how does an LED wall work" another supplier, ABM Srls, was ninth with a video uploaded 23 hours earlier: 7 views.
New videos get a chance. 7 of 79 positions went to videos less than 30 days old, four of them with fewer than 350 views. For "how does agrivoltaics work" a regional channel, AgriVolt FVG, was tenth with a 4-day-old video and 12 views.
So a well-titled upload is enough?
No. This is a one-day snapshot with no history: a signed-in viewer sees different pages, and we do not know how long those new videos will stay. But it fits what YouTube says: on niche searches what counts first is whether the video answers that question, and the big-view competitors are often on a related topic, not the same one.
Our control search made the point almost comically. For "algoritmo youtube", in the same test, 5 of the top 10 videos were "Super Algoritmo" wellbeing audio tracks from a single channel. The word matched, the topic did not.
Who owns the YouTube searches in your sector?
You have just seen a 34-view video in fourth place for "LED wall price". If yours is not there, a competitor's is. We check which of your customers' questions trigger a video block on Google and on YouTube, and who holds them today.
Book a strategy callWhy search matters more than suggested videos for a business (and where to check in YouTube Studio)
For a B2B company, search is the part of the algorithm it can influence: the question already exists, the customer types it. Suggested videos need a large audience of similar viewers for the system to learn from, and a market of a few hundred buyers does not supply one.
You check this in YouTube Studio. The Help page Understand your YouTube video reach lists the traffic sources, including Browse features (home and subscriptions), Suggested videos, Shorts, YouTube search and External, meaning websites that embed or link to the video. They sit under Content, the video's Analytics, Reach tab.
For a technical business video the two lines that matter are YouTube search and External. If a 300-view video gets 200 of them from search on a product query, it is doing its job. The reports are covered in our guide to YouTube Analytics.
External traffic depends on one choice that is entirely yours: embed the video on the page of your site that answers the same question. It is also what video SEO for the page asks for.
What YouTube says about the most repeated myths (tags, frequency, edited titles)
Many of the rules you read about the algorithm have a written answer in YouTube Help. Side by side:
| What you read | What YouTube says | Source |
|---|---|---|
| Tags help rankings | Tags "are primarily used to help correct for common spelling mistakes" | Performance FAQ, YouTube Help |
| You must upload every day | "growth in views across uploads is not correlated with time between uploads" | Performance FAQ, YouTube Help |
| Changing a title or thumbnail resets a video | It can be an effective way to get more views; the systems respond to how viewers interact with the video | Performance FAQ, YouTube Help |
| You can pay to rank in search | YouTube does not accept payment for organic placement | How YouTube Search works |
| 70% of watch time comes from recommendations | A 2018 statement by Neal Mohan at CES; the 2021 official post says "more than subscriptions or search", with no percentage | Tubefilter 2018; YouTube blog 2021 |
| The algorithm rewards CTR | The 2016 paper ranks by expected watch time per impression, precisely so as not to reward videos people abandon | Covington, Adams, Sargin, RecSys 2016 |
The last row has a practical consequence for thumbnails: one that promises what the video does not deliver lifts clicks and cuts watch time, the signal that counts. How to build an honest one is in our guide to YouTube thumbnails; what CTR really measures we have explained for Google.
The algorithm and AI answers (where a video gets cited)
On Google, "youtube algorithm" already has an AI answer. On 13 October 2026, on mobile in the UK, the AI Overview cited 8 sources, 6 of them on youtube.com, with a carousel of 3 videos further down. In Italy, for the equivalent search, it cited 10 sources, 5 on youtube.com. The term gets about 480 searches a month in the UK (Google Ads Keyword Planner via DataForSEO, 12-month average).
The AI definitions lean on the creator version: a matchmaker that keeps people watching. YouTube's own documents talk about valued watch time and satisfaction surveys.
For a business the lesson is practical. A video becomes a source when the engine can tell what it covers and which question it answers: a title that matches the question, a description that sums up the answer, chapters and a clean transcript. Those are the same elements that feed relevance in YouTube search, and none of them guarantees a citation. What we measured across 24 searches is in YouTube videos in AI Overviews and ChatGPT; how we work on presence in AI answers is our GEO service.
Five moves for a business channel (built for search, not virality)
1. One video for each pre-purchase question
Start from the questions your sales team hears every week, such as "which pixel pitch do I need for an indoor LED wall". Each question is a video, with the question in the title.
2. A description and chapters that state the answer
Relevance is read from the description and content too. Two sentences of answer and chapters for each step help anyone, person or AI, understand the video without watching it.
3. An honest thumbnail
Promise only what the video shows. A click that ends after ten seconds is exactly what Google's paper chose not to reward.
4. Measure by traffic source
Check YouTube search and External for each video. Total views on a business channel say little: 300 views from buyers are worth more than 30,000 from the curious.
5. Push the videos that answer
In a small market, recommendation starts slowly. You can speed it up by embedding videos on your site and with YouTube ads aimed at past site visitors. How to set the channel up first is in YouTube channel for business; for growing numbers without buying them, see how to increase YouTube views.
We have done the same on Google for years with SEO. For Sunscape, which looks for land for agrivoltaic plants, organic impressions went from 329 in March to 5,809 in August 2026 by answering landowners' questions one by one. On YouTube the logic is the same; only the format changes.
Frequently asked questions about the YouTube algorithm
No, not directly. YouTube Help lists relevance, engagement and quality for search and does not mention subscriber count. In our test on 8 searches, 12 of 79 first-page videos had fewer than 1,000 views.
No, hardly. YouTube's performance FAQ says tags are primarily used to help correct common spelling mistakes. Title, description and the video itself carry more weight.
No. YouTube says growth in views across uploads is not correlated with the time between uploads, and recommends focusing on quality.
No. YouTube says it does not accept payment for placement in organic search results. You can pay for ads, which appear in separate slots.
It depends: the principles are the same, but Shorts are a separate traffic source in Analytics, and YouTube gives more weight to relative watch time for short videos and to absolute watch time for long ones.
It depends: according to YouTube Help, viewers subscribe to dozens of channels on average and may not return for every upload. A subscription is one signal among many, not a promise of delivery.
In short (and the 300-view video)
Two systems: search and recommendation.
Search looks at relevance, engagement for the query and quality.
Recommendation looks at clicks, watch time and star surveys.
Across 8 sector searches, the top video had a median of 23 times fewer views than the most watched.
Tags and daily uploads: YouTube says they count for little.
For a business, the algorithm worth winning is the one that answers questions, not the one that makes things go viral. The right video in a niche market might get 300 views in a year. If 200 of them come from someone choosing a supplier, it is the most profitable video on the channel, and no creator guide would count it as a success.
