Direct answer: create a Language evidence card for every useful public phrase. Record the Verbatim phrase, Public source, date, surrounding context, question or audience job, repetition, and an Evidence note. Group phrases only after capture, preserve contradictory examples, and apply the rule No demographic inference from wording alone.
- Source: public URL or durable reference, source type, publication date and access date.
- Language: short verbatim phrase, surrounding context and apparent question.
- Interpretation: provisional audience job, theme candidate and contradictory example.
- Quality: repetition key, evidence note, uncertainty label and reviewer.
This worksheet records public wording. It does not create a brand voice, write captions, generate DM scripts, profile private individuals, infer demographics, or explain automated bulk collection. The output is a small research corpus for understanding how public audiences describe a niche and the questions they ask.
Use the results as one evidence input to the niche validation scorecard. A phrase appearing several times in a selected public sample can justify a research question; it does not establish the size or purchasing intent of a wider market.
Write a bounded research question
Define the knowledge gap before collecting language. Examples include: which words people use to distinguish two adjacent styles; what beginners ask before understanding a process; or which outcome they describe when praising a format. Name the niche, public surfaces, geography if relevant, language, and collection window.
Avoid starting with “find words that sell.” That objective encourages selective capture and turns observations into copy before they are understood. Prefer a question about meaning, confusion, comparison, sequence, or desired progress.
Set a stop point such as a date or number of reviewed public items. The worksheet is a manual research aid, not an instruction to collect at unlimited scale. Record skipped sources and why they were outside the boundary.
Select Public source types deliberately
Choose public pages where the relevant audience naturally discusses the subject: public comments, public community questions, search suggestions, public reviews of related tools or formats, and creator-owned public replies. Follow each source's current access rules and use ordinary manual review rather than automated collection.
Record the URL or durable reference, publication date when visible, access date, and context. Do not copy usernames, avatars, profile descriptions, or unrelated personal details into the working sheet. A short phrase and sufficient context are usually enough.
Keep source types separate during review. Search wording, a forum question, and a public comment arise in different contexts. Similar words across them can be interesting, but should not erase those differences.
Complete one Language evidence card per observation
Fields: card ID; research question; Verbatim phrase; Public source type; source reference; access date; surrounding sentence or context summary; apparent question; possible Audience job; theme candidate; repetition key; Evidence note; and reviewer. Use one observation per card.
The verbatim field should preserve spelling and phrasing when the excerpt is short and necessary for analysis. The context summary explains what the person was responding to. Without context, the same phrase can mean approval, sarcasm, confusion, or comparison.
Keep the possible job provisional. “How long did that take?” may concern effort, process, or simple curiosity. Record the ambiguity rather than choosing the most commercially convenient interpretation.
Capture exact wording before normalising it
Do not replace a public phrase with internal terminology during collection. If people repeatedly use “setup,” while the team says “pre-production,” both can be recorded, but the audience wording remains visible. This helps later content use recognisable language without pretending every phrase has identical meaning.
Add a normalised key in a separate column only for grouping spelling variants or clearly equivalent forms. Document the grouping rule. Do not merge adjacent concepts merely because they share one word.
Preserve questions as questions. Their form reveals what information is missing: “what,” “how,” “which,” “when,” and comparison questions can imply different answer jobs. Do not prematurely convert them into headlines or sales prompts.
Group themes after enough cards exist
Review cards in a batch and cluster similar observations. Name each group with a neutral description, then list included and excluded examples. Count cards, distinct source items, and source types. Several phrases on one page are not the same as repetition across several independent contexts.
Keep outliers and contradictions. A phrase that appears once may expose an important ambiguity, while a popular term may be used with conflicting meanings. The group summary should state both the pattern and what remains uncertain.
Have a second reviewer inspect a small sample. Discuss disagreements about meaning and context. Update group definitions prospectively rather than silently moving old cards until the desired pattern appears.
Apply No demographic inference
No demographic inference: do not infer age, gender, location, income, identity, or private characteristics from a phrase, username, writing style, or topic. The worksheet captures language in a public context, not a profile of the person who used it.
A geographic Trends filter describes where aggregated search interest was measured, not the demographic identity of an individual phrase author. Keep aggregate context in its own field and never attach it to a person.
If the research question genuinely requires audience characteristics, use an appropriate voluntary research method with explicit fields. Do not make the language worksheet perform a job it was not designed to do.
Use Google Trends with explicit settings
Google Trends can provide supporting context for terms. Record exact spelling, comparison term, geography, time range, category, search surface, and whether the item is a term or topic. A topic can group related concepts while a quoted phrase can narrow to exact wording; the distinction matters.
Relative search interest is not a count of subscribers and does not prove offer demand. Use it to see whether two public terms behave differently enough to warrant more research. If data is insufficient, record that state.
Do not swap in a nearby high-interest term after seeing the result without creating a new comparison record. Post-hoc changes make the research path difficult to audit.
Translate language into answer questions, not copy
For each well-supported theme, write an answer question: what does the audience need clarified, compared, demonstrated, or sequenced? Attach supporting card IDs and one contrary example. This creates a trace from public language to a content research task.
Move approved questions to the content strategy planning guide or the content ideas guide. Those pages handle planning. This worksheet does not write finished content or decide its sales role.
Keep exact public phrasing as evidence, not a phrase bank to paste into messages. Original content should answer the underlying need in the creator's own accurate voice.
Run privacy and quality review
Before sharing, remove personal identifiers, unnecessary sensitive context, private messages, and long copied passages. Confirm every card came from a public source within the declared window and contains enough reference for an authorised reviewer to verify.
Check that theme counts reconcile to cards, source diversity is visible, contradictory examples remain, normalisation rules are documented, and no demographic field has been inferred. Mark uncertain interpretation as uncertain.
Archive the collection version with its research question. Later language can be compared through a new window without rewriting the earlier snapshot.
Publish a research summary with traceable counts
Summarise the collection window, number of cards, number of distinct source items, source-type mix, theme definitions, uncertain cards, and contradictory examples. For each finding, cite the supporting card IDs. A finding should say “within this public sample” rather than imply that the wording belongs to every person interested in the niche.
Separate observation, interpretation, and action. Observation records that five distinct public questions used a comparison between two formats. Interpretation suggests the audience may need a clearer distinction. Action proposes one explanatory content test. Keeping all three lines visible lets a reviewer challenge the interpretation without losing the underlying records.
Record rejected findings too. A phrase may be memorable but appear only once, come from an unrelated context, or depend on a meaning the team cannot verify. Marking it rejected prevents the same weak idea from returning later without its history.
Give the summary an expiry or refresh date when the language is time-sensitive. Public vocabulary can change, and a term can acquire a different meaning in another community. A dated snapshot remains useful as historical evidence even when a fresh window is needed.
Limitations and sources
Keep a short methods note beside any downstream brief. Include the research question, collection window, public source types, card count, grouping rule, and known blind spots. This prevents a phrase from travelling into planning after its context has disappeared. When the source set is too narrow to support a theme, label the item exploratory and collect a cleaner sample before acting on it.
Limitations: a selected public sample can overrepresent vocal participants, platform-specific conventions, or current events. Repetition inside the sample is not population prevalence, and language does not by itself reveal intent or identity.
The method adapts GOV.UK guidance to record observations and group themes and use evidence-led language. Term caveats come from Google Trends Help. The card and no-inference review are original SirenCY editorial tools.