Lumen by Talkwalker Foundations
Lumen by Talkwalker is the listening and analytics layer. This domain tests three things: whether you can write a query that returns the right conversation, whether you know what each source actually covers, and whether you can read the resulting metrics without over-claiming.
Query construction
- The Boolean core: AND narrows, OR broadens, NOT excludes. Quotes lock a phrase, parentheses group logic, wildcards catch spelling variants.
- A brand query is a taxonomy, not a keyword. Official name + handles + hashtags + common misspellings, then context terms to disambiguate, then exclusions for the biggest noise clusters.
- Test before you trust: run the query, sample around 50 recent mentions by eye, and estimate what share are on-topic. A query nobody validated is a guess with a chart attached.
Coverage and metric meaning
- Sources typically include X, Instagram, Reddit, forums, blogs, news and review sites. Each has different volume and different access, so cross-source comparisons are directional, not precise.
- Volume, reach, sentiment and share of voice are all computed from the matched mention set. Change the query and every number moves - which is why query changes must be versioned.
- Sentiment is model-assigned per language. Sarcasm, slang, code-switching and quoted criticism are the standard failure cases.
Worked example
A drinks brand named Juniper wants earned conversation only. Query: ("Juniper" OR "JuniperDrink" OR #juniperdrink) AND (soda OR tonic OR flavour OR flavor) NOT (gin OR berry OR recipe) NOT from:juniper. Reasoning: the OR block captures naming variants; the AND block removes the plant, the spirit and recipe content; the NOT clauses remove the largest noise clusters and the brand's own handle so owned posts do not inflate earned sentiment.
Traps
- Writing AND where OR belongs, collapsing volume to almost nothing.
- Excluding a term that is also a legitimate product name.
- Comparing a brand's 30-day window against a competitor's 90-day window.
- Reading share of voice as market share.
- Treating a sentiment percentage as survey evidence rather than a model estimate.