We Analyzed 143,498 Tweets. Here Is What We Found

By Michal Mazurek

We analyzed 143,498 saved X/Twitter matches from Syften's mention archive. After removing general keyword filters, handle-only filters, duplicates, and retweets, we were left with 70,125 original tweets that matched brand and product searches. 31.2% of those tweets had no account handle anywhere in the tweet.

That is the problem with monitoring only @brand mentions. People often talk about companies and products by name, by domain, through screenshots, in comparisons, or inside support complaints without tagging the account. If your monitoring starts and ends with the handle, those conversations never reach you.

This article explains what we counted, what we left out, and what to monitor instead. If you want the setup walkthrough after reading the data, see how to find and track mentions on Twitter.

What we measured

We started with 143,498 saved X/Twitter matches from Syften's mention archive. This article only uses totals. We are not publishing customer names, filter text, tweet text, account names, or individual matches.

First, we separated searches for a specific company or product from broad keyword searches. Exact product names, account handles, and company domains stayed in the analysis. Generic category, job, technology, and problem phrases were treated as general. When a filter could go either way, we left it out.

We also removed brand filters that only searched for an account handle. A filter that only searches for an account handle cannot find untagged mentions, so including it would skew the result toward tagged tweets and understate the problem.

Then we deduplicated by tweet URL and removed retweets that started with RT @. That left 70,125 original tweets from brand and product searches.

StepTweetsWhy it matters
All saved X/Twitter matches143,498Raw archive export before cleanup.
Unique tweets from brand searches77,669Deduplicated tweets after broad phrases and handle-only filters were removed.
Retweets starting with RT @7,544Removed so repeated shares did not inflate the result.
Original tweets left70,125What the 31.2% figure is based on.

The result: 31.2% had no account handle at all

Of the 70,125 original tweets left after cleanup, 21,857 had no account handle anywhere in the tweet. That is 31.17%.

BucketTweetsShare
No account handle anywhere21,85731.17%
Starts with account handle39,00155.62%
Has account handle later9,26713.21%

The important number is the first one. If a tweet has no account handle at all, then a handle-only monitor cannot find it. It does not matter how good the alerting, inbox, or dashboard is; the query is too narrow.

We are not claiming that every tweet starting with @someone is missed by handle monitoring. Some of those tweets do tag the brand. Many are replies to customers, founders, journalists, or unrelated accounts. Without mapping every filter to every official brand handle, treating that whole group as "missed" would be too loose. So we are making the narrower claim: at least 31.2% of the original tweets in this analysis had no account handle at all.

What untagged brand mentions look like

Untagged mentions are not exotic. They look like ordinary people talking naturally. In pattern form, they often look like this:

  • A developer says they want to recreate a feature from a specific product.
  • A user compares three tools by name without tagging any official accounts.
  • A news account links to a funding story and names the company in plain text.
  • A customer mentions a pricing change, outage, or bonus without trying to contact support.
  • A job post mentions implementation experience with a specific vendor.

These are exactly the kinds of mentions a founder, support lead, marketer, or sales team might care about. They include comparisons, hiring signals, product feedback, category research, and competitor context. But none of them require the author to know or use the official account handle.

Why people do not tag brands

People do not write tweets for your alert setup. They write the shortest thing that makes sense to their audience.

  • They remember the product name, not the handle.
  • They are comparing several tools and do not want to notify every vendor.
  • They are replying to another person, not trying to reach the company.
  • They mention a domain, screenshot, article, repo, or feature instead of the account.
  • They are complaining, researching, or recommending something casually.

This is why @yourhandle is a useful search, but a bad complete monitoring strategy. It catches people who intentionally tag you. It misses people who simply talk about you.

What to monitor instead of only your handle

Start with your handle, but add the words people use when they are not trying to notify you. A practical brand-monitoring query usually includes:

  • Your official handle: @yourhandle
  • Your brand name: "Your Brand"
  • Your product names, feature names, and common abbreviations.
  • Your domain: yourbrand.com
  • Common misspellings, old names, and spacing variants.
  • Competitor comparisons, for example "alternative to competitor".
  • Category pain phrases people use before they know your product exists.

For Syften, a simple X/Twitter filter starts with the account, the brand name, and the domain:

@syften_com OR
(syften lang:en) OR
syften.com
-from:@syften_com

The brand-name part matters. If someone says "Syften" without tagging @syften_com, a handle-only search misses it. If someone links to syften.com without mentioning the account, a handle-only search misses that too.

How Syften handles this

Syften treats handle monitoring as one part of keyword monitoring, not the whole job. You can monitor handles, brand names, domains, competitor names, author accounts, and buying-intent phrases in the same place. Matches can go to email, Slack, RSS, API, or webhooks.

That matters because a useful mention is usually not a dashboard metric. It is a thread someone can answer, a complaint support can route, a comparison sales can learn from, or a competitor mention that explains how the market talks.

If the raw keyword filter is still too noisy, Syften can also apply AI filtering after the match. For example, a filter can match a broad brand or competitor name first, then use an $accept rule to keep only posts where the person is genuinely discussing the product, asking for alternatives, or comparing tools.

For the practical setup, start with the X/Twitter monitoring page or read the guide to constructing a good filter.

Limitations

This is not a universal measurement of all X/Twitter brand conversation. It is a measurement from Syften's archived tweet matches. The brands skew toward software and technical products, and the archive only includes tweets that matched configured filters.

We were conservative about which filters counted. We kept a filter only when it clearly tracked a company, product, domain, handle, or recognizable product name. Ambiguous filters were treated as general phrases and left out of the final count.

We also avoided claiming that every tweet starting with @someone is missed when you monitor only your brand handle. Some are direct brand tags. Some tag a person or unrelated account while mentioning the brand in plain text. That would require mapping every brand to every official handle and checking each tweet against that map. For this article, tweets with no handle at all give us the safer number.

The takeaway

Monitoring your handle is necessary, but it is not enough. In this analysis, 31.2% of original tweets from brand and product searches had no account handle at all. Those tweets would be invisible to a monitor that only searched for @brand.

The fix is straightforward: monitor the handle, the brand name, product names, domains, and the phrases people use when comparing, recommending, or complaining about tools in your category. That is how you find the conversations people did not deliberately send to you.

Michal Mazurek

Article by

Michal Mazurek

Michal Mazurek is the Founder of Syften. Michal has 7 years of experience helping companies set up social listening profiles that find useful conversations instead of noise. He's also a passionate engineer with 26 years of experience as a low-level programmer, web developer, security analyst, embedded developer, and sysadmin, including work with supercomputers.

Last month on Syften

Loading recent community coverage...

See who's talking about you

Search recent public conversations for your company or a competitor.

https://
Searching...