Has Douyin’s Algorithm Changed? Completion Rate May Matter Less Than You Think

On Douyin, the question is not only whether people finish a video, but what they can do after watching it. That does not make completion rate obsolete, nor does it mean saves have replaced every other metric. For tutorials, educational content, and decision guides, a save may indicate an intention to use the information later. For entertainment, short narratives, or product demonstrations, viewing quality, sharing, and action also matter. The practical shift is from completion rate alone to viewing quality, usefulness, and business outcomes. The question in the headline is not an announcement of a new ranking formula. This article concerns Douyin in China, not a claim about TikTok's algorithm.

Douyin's Safety and Trust Centre describes retrieval, multiple ranking stages, and several user behaviours used in scoring, including likes, saves, shares, and comments. This supports a multi-metric approach, not a universal hierarchy for every account, format, or video length. The official material checked for this article does not substantiate a platform-wide rule that saves now outweigh completion. Use published principles as context and your own content tests as decision evidence.

Completion helps diagnose whether a video holds attention, but length, traffic source, viewing intent, and structure affect it. A 15-second demonstration and a three-minute tutorial should not be judged against the same benchmark. Cutting essential steps may improve completion while making a tutorial less useful. Compare similar lengths, subjects, and traffic sources, then inspect average watch time, retention curves, and drop-off points where available. Early exits suggest checking the promise and audience fit; mid-video exits suggest checking repetition, pacing, or complexity. Use the definitions and fields actually available in your account.

A save may mean someone plans to follow the instructions later, lacks time to watch now, wants a reference, or simply collects resources. Likes, comments, and completion also have different motivations. It is inaccurate to treat saves as genuine and every other behaviour as superficial. For practical content, look for repeat viewing where measurable, specific questions, reports of use, or subsequent actions. If the account does not expose those data, mark them as unobserved. Do not convert save counts directly into loyal users or sales prospects.

‘Pay attention to analytics’ is an opinion. ‘Open the analytics area, find the video's available retention or viewing data, locate a drop-off, and review that exact segment’ is an executable path. A reusable tutorial should explain the audience, prerequisites, steps, success criteria, common mistakes, and exceptions. For opening design, do not stop at ‘hook people in three seconds’. Show the outcome, name the problem, and remove an introduction that does not help the task. These are editorial practices, not an algorithmic formula every video must follow.

A useful ending might be: ‘Use this checklist when you review your next video.’ If the content genuinely supports future reference, an optional save reminder can follow naturally. Deliver the tool and its use case instead of repeatedly asking for engagement. Do not hide the complete answer behind likes, comments, or follows, and do not offer fictitious rewards or growth guarantees. A viewer who never interacts should still receive the value promised by the headline.

After publication, classify comments into unclear steps, suitability questions, counterexamples or corrections, and business enquiries. Clarify confusing steps; turn recurring questions into an update or follow-up; correct errors openly instead of removing disagreement to manufacture approval. If pinning is available, use it for key steps and corrections, but do not make comments the only place to find essential information. Avoid coordinated scripts or account activity. Thoughtful replies help users and reveal needs; they do not guarantee another distribution push.

‘Be consistent’ can become a content calendar. ‘Measure ad conversions’ can become a diagnostic checklist from clicks through enquiries, qualified leads, and sales. ‘Make production professional’ can become a pre-shoot check for sound, lighting, captions, and demonstration. Checklists, steps, templates, tutorials, and comparisons reduce execution effort; they do not inherently earn higher algorithmic weight. A template needs instructions and an example, a tutorial needs failure conditions, and a comparison needs a consistent method. More information is not automatically more useful.

Tutorials and checklists can prioritise saves, specific questions, or reported use, with viewing quality as a guardrail. Entertainment and stories should focus on viewing experience, sharing, and audience response rather than forced templates. Product demonstrations and service explainers should continue to profile visits, compliant enquiries, qualified leads, and sales. Choose one primary objective and a few guardrails for each content job. If calculating saves per thousand plays, keep the window, denominator, and traffic sources consistent and do not call it a unique-user save rate. Leave unavailable repeat-visit or attribution data explicitly unknown.

This is a suggested test cadence, not an official learning period. Review the last 10–20 comparable posts and group them by length, audience, and organic or paid source. Choose three real questions and test an explanation-led version against a steps-or-tool version. Keep presentation, duration range, and observation windows reasonably similar without uploading identical videos. Record the topic, publish time, duration, sources, viewing, saves, specific questions, and enquiry quality. Publication order and audience mix remain confounders, so treat this as directional evidence rather than a randomised experiment. Extend observation when the sample is insufficient.

First, check whether the people saving are potential customers rather than peers or people who only need a free resource. Then examine whether the problem addressed connects to the actual service, whether the profile explains the offer and evidence clearly, and whether the enquiry path is convenient and follows platform rules. Check follow-up, buying readiness, and whether the conversion cycle has completed. WEPR recommends evaluating content research, query mapping, brand evidence, enquiry handling, and measurement as one business journey. The useful question is not whether saves beat completion, but whether the content helps the right people solve a problem and make a better next decision.