The real sentiment problem: why brands can no longer trust what they see online

Somewhere in the last eighteen months, a quiet line was crossed. The volume of synthetic activity online — bot accounts, AI-generated reviews, coordinated inauthentic engagement — grew fast enough that a lot of what looks like customer sentiment is no longer coming from customers at all. For brands relying on social to understand how they are perceived, that is not a small problem. It is a foundational one.

Most monitoring agencies are still selling a version of "we watch everything." The more honest pitch, and the more valuable one, is starting to be: we tell you what is actually real.

The shape of the problem

Synthetic sentiment shows up in three broad ways, and all three are getting harder to spot.

AI-generated reviews

Reviews that read fluently, mention the right product features, and hit a plausible star rating. Written by a model, posted by a network, sometimes to lift a brand, sometimes to drag a competitor. Platform trust and safety teams catch a share of them. They do not catch all of them, and the ones that get through are not obvious to the average reader.

Bot-driven engagement

Likes, shares, replies and follows generated by automated accounts. The technology is not new. What is new is how convincingly the accounts hold up under a casual scan — real-looking profile photos, plausible bios, activity patterns that resemble a person having a slow week.

Coordinated inauthentic conversation

Small groups of accounts, sometimes human-run, sometimes hybrid, pushing a narrative in a way that mimics organic momentum. A brand's social team sees a topic trending against them and reasonably concludes the customer base is upset. Sometimes that is true. Sometimes it is fifty accounts working a hashtag for a week.

The result for brands is that raw sentiment numbers have quietly become less trustworthy. A rising positive tide might be genuine advocacy or a promotional network. A wave of negativity might be a real problem or a coordinated push. Both look the same in a dashboard.

Why this is a differentiation problem, not just a data problem

For a monitoring agency, this changes the value proposition in a way most of the industry has not adjusted to yet.

The old pitch: "We will watch every channel and give you the full picture."

The new pitch: "We will watch every channel and tell you which part of the picture is actually customers."

Those are not the same service. The first is a volume play. The second requires:

• Human judgement on the floor, not just automated classifiers. Trained community managers who have spent enough time in the same brand conversations to notice when something is off.
• Cross-signal verification, so a spike is checked against ownership patterns, account age, engagement shape and posting cadence before it is escalated as customer sentiment.
• The discipline to filter out what does not belong, even when the raw numbers make the report look more impressive.

Most tools cannot do this on their own. Most agencies are not set up to do it at all.

What "real" looks like in practice

A few things separate a signal that reflects actual customers from one that does not.

• The conversation is spread across a mix of established accounts with real posting histories, not clusters of new or thinly populated ones.
• The language varies. Real customers repeat themselves in messy, human ways. Synthetic content tends to hit the same beats a little too cleanly.
• The engagement pattern includes friction — people arguing, disagreeing, asking clarifying questions — rather than a smooth wave of agreement.
• The complaint or praise is anchored to a specific, verifiable experience, not a generic feature description that could apply to any brand in the category.

None of these are absolute tests. All of them, applied together by a team that knows the brand, produce a much more reliable read than a raw sentiment score.

Why brands should be asking harder questions

If you are a brand paying for social intelligence, the useful question to put to your provider is not "how much are you capturing?" It is "how much of what you are capturing is actually customers, and how do you know?"

The answers you want to hear involve human review, verification steps, and a clear methodology for excluding inauthentic activity from the numbers that reach your leadership team. If the answer is a confident wave at a dashboard, the numbers you are seeing are probably softer than they look.

Where Burrow sits on this

Burrow's model was built around trained community managers on the floor of an Australian-based command centre, working alongside tooling rather than being replaced by it. That structure is what makes the "what is actually real" question answerable in the first place. Our team already:

• reviews flagged content before it feeds into client reports
• checks engagement patterns against account behaviour, not just keywords
• separates coordinated activity from genuine community sentiment in escalations
• gives clients a read on the underlying conversation, not just the raw metric

Final thought

Almost nobody in the monitoring space is talking about this yet, which means most brands are still receiving reports where synthetic and real sentiment are quietly blended together. That will not last. The agencies that get in front of it now — and the brands that start asking about it now — will be the ones who still trust their own numbers in two years.