Social media monitoring keywords: writing the list
Too broad and the team drowns in noise; too narrow and key posts never appear. How to build social media monitoring keywords properly.
Short answer
To write monitoring keywords properly, split them into five lists — brand, products, category, competitors and risk words — and make three decisions for each term: which match mode (word start, whole word or exact phrase), which context word it must appear with if it has other meanings, and who added it and why. A good keyword list is not long but precise: every term is tested before it is added, and terms that bring noise are pruned every month.
Why keywords decide everything
Monitoring only finds what it searches for. If a keyword is too broad, the team reads hundreds of irrelevant posts a day and burns out fast. If it is too narrow, important complaints and buyer questions never appear. Either way the problem is not the tool but the list. So the keyword list is not a document written once and forgotten — it is monitoring's main tuning tool and needs an owner.
Five lists
- Brand — the company name, abbreviation, spelling variants, domain and account handles.
- Products — product and service names; those made of common words go with context.
- Category — phrases that carry buying intent without a brand name: "English course for children", "office air conditioning".
- Competitors — main competitors' names and variants; a separate list so results do not mix.
- Risk words — safety, fraud, lawsuit, poisoning and so on, searched together with the brand name.
Keeping the lists separate keeps the results separate too: brand results go to the daily check, category results to lead and content work, competitor results to the monthly report.
Choosing the match mode
Azerbaijani is agglutinative — suffixes attach to words: "Bravo", "Bravoda", "Bravonun". So in most cases matching by word start is the most useful — one query catches every form. But for short, common words that brings noise: a brand called "Ay" would also catch "ayaqqabı" (shoes) and "aylıq" (monthly). Such words get whole-word mode, and multi-word names get exact-phrase mode.
- Word startLong, distinctive names and category words: catches suffixed forms.
- Whole wordShort names and names that coincide with common words: only the word itself.
- Exact phraseMulti-word names and phrases: "Gülər Şirniyyat", "how to choose".
Context for ambiguous words
If a brand name coincides with a common word or a personal name, do not search it alone. Pair it with context words: "Gülər" + "cake", "sweets", "order"; "Bravo" + "market", "branch", "discount". Also record the name's other meanings in the brand profile so whoever — or whatever — reads the post can tell them apart. Context words raise precision but lower coverage: a post that mentions you without context may be missed. That is a conscious choice and should be noted.
A keyword register
- The term or phrase.
- List — brand, product, category, competitor, risk.
- Mode — word start, whole word, exact phrase; any context words.
- Why it was added — for which need.
- Precision note — what share of results was relevant at the last check.
- Owner and date of last review.
Testing before adding
- A one-week trialRun the new term in a separate search for a week.
- Count precisionOf the first 30–50 results, how many are really about you or your topic?
- DecideHigh precision — add it; medium — add a context word or change the mode; low — do not add it.
- Record itWrite the result in the register so the same term is not tested again a year later.
Monthly pruning
Once a month, review the register: which terms brought only noise this month, which found nothing, which new terms have started appearing in discussion. Narrow or remove noisy terms. A term that finds nothing is harmless but bloats the list — if it has found nothing for three months, remove it. And remember to remove campaign terms when a campaign ends.
Illustrative example
This is an illustrative example. An optician chain called "Nur" first searched for "Nur" by word start and got more than 300 results a day — the words "nur" (light), "nurani", the name "Nurlan" and hundreds of personal names. After building a register, "Nur" moved to whole-word mode and was paired with the context words "optics", "glasses", "lens" and "branch". Daily results fell to 20, of which 15 were genuinely about the chain. The register noted that a few context-free mentions might be missed.
Limitations
Even the best keyword list works only in the sources you watch and only on text: a logo in an image or a name spoken in a video is not found. There is always a trade-off between precision and coverage — you cannot maximise both. Tracking personal names as keywords, especially of people who are not public figures, creates privacy risk and is usually unnecessary.
Common mistakes
- Keeping all terms in one list.
- Searching short, common words by word start.
- Adding terms without testing.
- Keeping campaign terms after the campaign ends.
- Having no owner for the register.
Keywords in Vexvon Monitoring
In Vexvon Monitoring each company has its own keyword list and profile. The default mode is word start — to catch Azerbaijani suffixes; there is whole-word mode for short, common words and exact-phrase mode for multi-word names. Text is normalised for Azerbaijani (İ, ı, I and i are treated as one), a term found in a title is kept as the stronger signal, and link text is searched too. When the profile lists the name's other meanings, the AI sets those posts aside as "not about us". More on the Vexvon Monitoring page.
Next step
Split your current keywords into the five lists, choose a mode for each and write the first version of the register. Put terms with an empty "why it was added" column through a one-week trial. Other articles are in the monitoring queries, data and reliability section; to look at it together, get in touch.
- Monitoring: queries, data & reliability6 min readSocial media monitoring privacy and platform rules
- Monitoring: queries, data & reliability5 min readSocial listening vs monitoring: the real difference
- Monitoring: queries, data & reliability6 min readSocial media monitoring dashboard: metrics that matter