Capture product-name variants in an alias register: the exact wording, language, evidence of use, matching conditions and next review date. Start with official names and observed public mentions. Test each new nickname, misspelling or transliteration before adding it to your regular query. Keep generated spellings in a separate candidate list until you find evidence that people use them.
That is an editorial workflow recommendation. Platforms do not require an alias register. It follows the test-and-refine approach in X's query documentation and Sprout Social's guidance on maintaining listening queries.
Separate observed names from guesses
Give every alias a type. The distinction tells the next reviewer what needs checking.
- Official name. Confirm it against the product page or local packaging. Record the model or version it identifies.
- Observed nickname. Preserve the wording from a relevant public post. Explain what connects the nickname to the product.
- Observed misspelling. Record the spelling as written. Do not silently correct it in the register.
- Transliteration. Record both the original script and the variant. Ask a reviewer who reads that language to check the context.
- Candidate. A colleague, language model or spelling tool suggested it, but you have not confirmed public use.
A nickname can refer to an entire product family. A translated category word can refer to competing products. Neither belongs in a product-specific query without a condition that resolves the ambiguity.
ICU's transliteration documentation explains how characters move between scripts without translating the underlying words. It also describes ambiguity and differences between transliteration methods. A mechanically generated spelling therefore needs a usage check before you treat it as audience vocabulary.
For consistent decisions across languages, use a multilingual listening codebook to record how reviewers identify the product and handle unclear references.
Keep evidence beside each alias
Start with public posts that clearly identify the product through its full name, an official link or enough surrounding context. Look for alternate wording in those posts and their accessible discussion. Search each candidate separately to see whether it refers to your product elsewhere.
For each real example, record the public URL, post timestamp, access date and a short context note. Record a language as uncertain when you cannot identify it confidently. The script alone does not establish the writer's country.
Use the following register structure. All names, source IDs, observations and dates in this table are hypothetical. Nemi Loop is an invented reusable-bottle example; these rows report no actual customer behavior.
| Alias and type | Language/script | Evidence note | Matching condition | Review date |
|---|---|---|---|---|
| Nemi Loop, official | English/Latin | Product page A, checked 2026-09-20 | Exact product name | 2026-10-20 |
| Nemi Lopp, misspelling | English/Latin | Synthetic post P1, 2026-09-21; product link confirms identity | Keep the full phrase | 2026-10-05 |
| little loop, nickname | English/Latin | Synthetic post P2, 2026-09-22; full name appears nearby | Require Nemi; unrelated loop discussions excluded | 2026-10-05 |
| ネミループ, transliteration | Japanese/Katakana | Synthetic post P3, 2026-09-23; context reviewed in Japanese | Test the whole string; exclude unrelated names only after review | 2026-10-05 |
In your working sheet, also store a stable product ID, alias owner, status, first and latest observed dates, last review date and query version. Keep the original wording alongside any normalized search form. That makes a later correction traceable.
Use statuses such as candidate, testing, active and retired. A retired row retains its evidence and retirement reason. Removing it from the live query should not erase the explanation for an older report.
Test what your search system already handles
Some apparent variants require no extra query terms. X documents case-insensitive search operators, so separate uppercase and lowercase versions add no case coverage there. Its search queries with accents match both accented and unaccented forms. Its filtered-stream rules handle accented keywords differently.
Record the endpoint or tool feature you tested. A successful search test does not establish that a live stream behaves the same way. Do not assume punctuation, word spacing, hashtag forms or transliteration receive equivalent treatment elsewhere.
Search coverage also limits what you can validate. X's Search Posts documentation describes keyword, hashtag, mention and URL searches. Recent Search covers the last seven days. Full-archive search has separate access requirements, and API use requires a developer account, a Project and App, and credentials.
A nickname absent from a recent search may occur outside that window. Record the window before calling the term unused. Use only sources your approved access can retrieve; an expanded keyword list does not create access to private material.
One further X-specific gap matters when checking examples: search matches a quote post's own content, rather than the original post it quotes. A product name visible in the quoted original may therefore fail your text query. Record that as a matching limitation instead of inventing another spelling.
Judge each addition by the posts it adds
Test a candidate against the same source set and time window as your existing query. Keep the baseline results, then identify which posts the candidate adds. Deduplicate by post ID before counting. For repeated copies of the same message, follow a duplicate-post review method so repetition does not masquerade as separate evidence.
Review a declared batch of added posts. Write down how you selected it, such as all available results in a stated window or a fixed batch in chronological order. Label each post relevant, irrelevant or unclear. Count unclear cases separately until a reviewer resolves them.
For resolved cases, precision is relevant results divided by reviewed results. This follows the information-retrieval definition of precision.
The following results are hypothetical and contain no unclear cases. Both rules use the same source set and time window. All 40 posts added by the broad rule are reviewed; the narrower rule returns a subset of those posts.
| Candidate rule | Added posts reviewed | Relevant | Precision in this batch |
|---|---|---|---|
| little loop alone | 40 | 12 | 12 / 40 = 30% |
| little loop with Nemi | 12 | 10 | 10 / 12 = 83.3% |
The narrower rule retains 10 of the broad set's 12 relevant posts and loses two. It gives the reviewer fewer irrelevant posts, but misses genuine mentions without the brand name in this example. Each precision denominator is the number of posts returned by that rule. Keep known relevant examples and rerun them after every exclusion change to catch that loss.
Choose the rule according to the work. An exploratory review can tolerate more irrelevant results than an alert sent directly to a response team. There is no universal precision threshold for an alias. For ambiguous names, design exclusion rules around the irrelevant meanings, then check that those rules preserve your known relevant examples.
These batches do not measure how common a nickname is among customers. They also cannot establish whole-platform recall. Recall needs a denominator of relevant items, including those your query missed. Report the source set, window and selection method with the result.
Set the next review when you activate the alias
Assign an owner and a calendar date before moving a tested alias to active. As a starting recommendation, review new aliases after two weeks and stable aliases monthly. Shorten that interval during a launch or a sudden rise in irrelevant matches.
At review, decide whether to keep the rule, add context, return it to testing or retire it. Check both new results and previously confirmed mentions. Save the changed query with its effective date so a rise in captured mentions is not automatically read as a rise in conversation.
Open your current keyword list and select one ambiguous alias. Add its evidence link, language, matching condition and next review date. Then inspect the posts it contributes before adding another spelling.
