Start Xiaohongshu Diandian AI optimisation by identifying the user's question, the entry point where the brand appears and whether the answer is accurate. Then decide what information to add. It is not repeating a brand name in notes or buying a fixed list position.
Brands can first answer real questions and correct product information, then observe results under consistent conditions. An operating team cannot guarantee whether an answer uses particular content or how brands are ordered.
This guide observes the Diandian or related AI answers and recommendation lists that a brand actually sees. Record different entry points, accounts and scenarios separately; do not assume they form one permanent ranking. This is a content and measurement method, not an official Xiaohongshu algorithm-weight specification.
Identify the brand's actual problem
Do not increase publishing immediately because a competitor appears. Save the question, answer, entry point and time, then classify the problem.
Observation / Check first / Possible action
- Brand absent from a relevant answer;Check first:Does the question fit the business, and is specific, public, accurate information available?;Possible action:Answer underserved scenario questions rather than duplicating brand introductions
- Brand appears with incorrect product or store information;Check first:Do the website, profile, notes and historical materials conflict?;Possible action:Align facts first, then update or report verifiable errors
- Only competitors appear;Check first:What conditions, evidence and questions are covered in their descriptions?;Possible action:Identify genuine differentiation and information gaps; do not disparage rivals or fabricate reviews
- Brand mentioned without enquiries;Check first:Are suitability, service explanations and the in-platform consultation path clear?;Possible action:Improve the next step; do not assume AI exposure is either worthless or sufficiently targeted
The table is a diagnostic starting point, not a causal conclusion. Brand absence may also reflect entry-point changes, personalisation or that particular answer. One screenshot does not establish an account restriction.
Build a question set you can retest
Collect questions from customer enquiries, support records, comments, search suggestions and sales interviews. Remove personal information before organising the wording.
Describing audience, task, scenario and concern is more useful than collecting product names alone. Corporate-event photography provides the following illustration.
Question type / Example
- Scenario;Example:How should a company choose event photography in Shanghai?
- Audience;Example:What photography setup does a small team need for a sharing event?
- Delivery;Example:What photo-delivery terms should be agreed after an event?
- Comparison;Example:How does conference coverage differ from brand-event photography?
- Risk;Example:Which services are easy to overlook in an event-photography quote?
- Brand verification;Example:What should be checked when assessing a provider for a multi-venue event?
These demonstrate question structures, not verified search volumes or actual client projects.
For each question, record the audience, intended task, facts the brand can support, existing notes and missing evidence. Explore new questions separately; do not mix exploratory results directly into the original trend comparison.
Align brand facts before scheduling content
Use a brand-fact register to check accuracy and consistency.
- Brand and legal-entity names, common abbreviations and former names.
- What the product or service does, and suitable and unsuitable circumstances.
- Location, opening or service coverage, verifiable contact information and official entry points.
- Price conditions, specifications, service inclusions and validity dates.
- Cases, images and qualifications that may be used publicly, with relevant permissions.
- The owner responsible for verification and updates, and the last check date.
Different notes can draw on the same facts without copying the entire register. When prices or services change, prioritise older descriptions users may still encounter.
Answer one specific decision question per piece
Begin with who the advice applies to and the answer, then give reasoning, actions, evidence and limits. Clear structure first helps users understand and decide; it should not be sold as a guaranteed AI-citation format.
The following fictional writing example demonstrates the information difference only.
An insufficient description
“Choose us for corporate-event photography: a professional team, beautiful photos and dependable service.”
A more useful description
“Before choosing corporate-event photography, specify event duration, venue count, essential moments and intended use of the images. Main-stage coverage alone cannot be staffed like an event also requiring guest portraits, exhibition-area coverage and short videos. A quote should confirm image quantities, retouching scope, delivery timing and usage rights.”
The second version makes no performance promise but gives usable decision criteria. A brand describing its own service should add real staffing, work samples and limits, not present this teaching example as an actual service commitment.
Give different notes distinct jobs
A brand can address the same broad need over time without repeatedly posting the same introduction under new titles.
Introductory content explains the category and basic decisions, such as what event photography includes.
Scenario content addresses a constraint, such as what to clarify first with a limited budget.
Comparison content explains conditions and trade-offs without portraying competitors as entirely flawed.
Evidence content presents authorised cases, working processes or verifiable materials with sample details and limits.
Service content explains what to prepare before an enquiry and the permitted in-platform next step.
First check whether existing content can accommodate missing sections. Consider a separate topic when the user task is genuinely different, not for every near-synonym.
Use comments to answer follow-ups, not manufacture evidence
Comments reveal omissions and let users obtain explanations. Repeated questions should inform the body or the next update.
Disclose brand and commercial relationships. Do not arrange fake users to ask and answer each other, copy positive reviews or use comments as an intermediary for concealed contact details. This article does not claim that comment count, length or response speed is a proven Diandian ranking weight.
Observe under consistent conditions, not through one screenshot
Keep the following records for each retest.
- The full question and test date.
- The app, page entry point and visible version information.
- Recordable conditions such as account, login state and region.
- Whether an answer suitable for evaluation was generated.
- Whether the brand appeared, was explicitly recommended and was described accurately.
- Sources or citations actually displayed; record unknown if none are shown.
- Competitors, answer changes and concurrent content or advertising changes.
Do not automatically record a mention as a citation of your note. Without a traceable displayed source, you can establish a mention, not infer the citation path.
You can calculate sample-level measures.
Sample mention rate = valid tests in which the brand appears ÷ all valid tests.
Define valid tests, repeat counts and treatment of failed or missing answers in advance and explain them. This is the team's observation metric, not an official platform-wide market share.
If an intervention combines many new notes, advertising and information updates, initially record any subsequent change as an association; do not attribute all of it to one action.
Separate visibility from business value
Seeing a brand does not necessarily produce an immediate click or enquiry; an enquiry may not fit the target customer.
Business records should distinguish sources too.
- Mentions, recommendations and factual accuracy in AI answers.
- Search entries, profile visits or other data the backend actually provides.
- Relevance, repeated questions and follow-up handling of in-platform enquiries.
- Business opportunities confirmed with sales under compliant, authorised conditions.
Some entry points do not provide full attribution. Keep missing data unknown rather than assigning leads to an AI answer they may have seen.
Starting a limited improvement cycle
Select questions directly relevant to the business, preserve a baseline and check existing notes and brand facts. Fill the clearest information gaps, log changes and retest under the same conditions.
If brand information was wrong, first check whether it is corrected. If a scenario was not covered, check whether the content now answers it. Do not reduce every review to a position number.
Frequently asked questions
Can a Diandian AI list position be guaranteed?
No. Content and information can be improved, but answers, displays and selection are not controlled by the brand. Scrutinise fixed positions or permanent-recommendation promises.
Should Diandian, question-answer features and regular search be combined?
Not directly. Name the entry points actually observed, retain their conditions and results separately, then assess comparability. Preserve records when product names or access points change.
Is it worthwhile without many followers?
You can first test question-content fit, but cannot promise that low-follower accounts will receive exposure. Account scale, content conditions and the current environment require observation.
Will repeating brand names and adding keywords help?
Brand names and topics should be clear and accurate, but mechanical repetition does not replace an answer. Add suitability, evidence and limits rather than targeting a fixed repetition count.
How often should performance be checked?
Set a regular cadence based on business changes and available resources. Update facts first when products or stores change materially. A retesting cycle is a work plan, not an official time-to-results promise.
Why are there mentions but no enquiries?
The question may be informational, users may be unsuitable, the service unclear or consultation inconvenient. Separate audience task from conversion-path problems; do not calculate revenue directly from mentions.
Should we buy comments or use an account network to push rankings?
Do not manufacture evidence through fake reviews, coordinated engagement or repeated content. Multiple accounts need real roles, distinct information and necessary disclosures; they must not evade rules.
Start with a specific question
Prepare common customer questions, observed answers, brand facts and existing notes. Then decide whether information governance, missing content or consultation handling needs attention first.
WEPR's Xiaohongshu marketing service can combine search, content and Diandian visibility in one execution plan. See the Diandian AI practical guide and AI-search measurement guide for further detail.
Source:Xiaohongshu marketing service
Source:Diandian AI practical guide
Source:AI-search measurement guide
