
A client's marketing lead put a question to me in a borrowed meeting room: her agency had proposed a virtual presenter for a skin-care line, the deck promised lower production costs and round-the-clock posting, and she wanted to know whether the audience would forgive her for it. The consumer data says most of them will not forgive a product claim that turns out to be synthetic, which is why the sensible move is not a yes or no but a set of review points before anything goes live. If your team is weighing ai automation for business in the same breath as an AI influencer pilot, the two decisions need separate owners.
As of November 2026, the public numbers are hard to argue with. A ResearchGate-published study found that only 26.4% of consumers would actively engage with an ad featuring an AI influencer, while 45.3% would view it without acting and 28.3% would ignore it outright. A Clutch survey adds the sharper finding: 71% of consumers would question a recommendation if a human influencer were found to use AI-generated content.
Three survey results carry most of the weight in this debate, and they measure different things.
The ResearchGate study asks how people would behave toward an ad. Most of the audience does nothing at all, which is a distribution problem rather than a trust verdict. The Clutch survey asks how people would feel once AI is discovered, and the answers are harsher. In the same survey, 53% of consumers said they were less likely to buy from a brand known for using AI in social content.
| Result | Figure | What it measures |
|---|---|---|
| Would view an AI influencer ad without acting | 45.3% | Attention without response |
| Would ignore the ad completely | 28.3% | Active rejection |
| Would actively engage | 26.4% | Clicks, searches, comments |
Read together, the two surveys describe a market where most people scroll past, and the ones who notice mind how they were told. A brand that posts AI content quietly is betting on nobody noticing; a brand that labels it is betting on the audience tolerating an honest synthetic presenter.

The Clutch results are worth stating in full because they separate the offence from the defence.
On the offence side, 42% of consumers view AI positively when it is used for educational content, but 64% would trust a brand or creator less if AI were used to misrepresent products. The line the audience draws falls on honesty, not on whether a face is real.
On the defence side, nearly 90% of consumers say it is important to disclose AI-generated content. That figure is the most useful planning input in the whole body of research: most buyers want the label, and legal review is only catching up with what they already expect.
The Clutch survey also found that 71% of consumers would question a recommendation if a human influencer were found to use AI-generated content. That specific wording matters for agency discussions, because it covers the mixed case, where a real creator uses AI to generate product visuals and never says so. When the audience finds out, the recommendation loses its weight, and the tool was never the thing they were trusting.
Two rules have already changed the operating environment, and both are now in force.
As of June 2026, New York requires certain ads featuring AI-generated human performers to include a disclosure. Transparency obligations covering AI-generated content have applied in the European Union since 2 August, 2026.
For a Singapore brand, neither rule binds directly. They bind through the market you sell into and through the platforms you buy on. A campaign shot for a US or European audience needs the label in the creative, not in the terms and conditions. The practical answer is one disclosure template per market, reviewed by legal before the brief is signed, and a named person who checks that the label survived the localisation.
The commercial case for AI influencers holds at the top of the market, where the numbers are public.
Aitana Lopez has over 400,000 followers on Instagram, and The Clueless, the Barcelona company behind her, has reportedly earned over $11,000 per month from that work. Euronews reported that the virtual model can earn up to €10,000 a month, with an average nearer three thousand euros, and that she earns just over €1,000 per advert. Earlier in her run, she had gained more than 343,000 followers in a year and a half.
Samsung's Sam shows the brand-owned version. Sam is openly presented as a virtual influencer, and the company reported that Sam's positive comments during Samsung Live in Mexico increased sales through the live portal by 30%. It is worth reading the source carefully: the figure is a company statement about a live-shopping portal, not an independent study of brand trust after the campaign.
The pattern in both cases is the same, and it is what a budget has to cover: the persona's identity is obvious and its role is clear. Neither Aitana nor Sam asks the audience to believe that a synthetic presenter personally tested a product in the way a human creator would.
The cost question is the one that starts most of these conversations, and Singapore teams tend to compare the wrong line items. The visible saving is production: no studio day, no reshoot when the creator changes the script. What the comparison usually leaves out is the oversight, the disclosure review per market, the measurement baseline and the cost of a stop-and-repair if the audience turns.
A fair comparison therefore runs on the whole programme, not on a single post. Ask what the human creator route would cost for the same number of posts, same channel, same audience segment, including the approval time you already pay for. Then add the new costs the AI route brings, which are review, legal wording and a benchmark study. When the gap is thin, the trust risk is being taken for very little, which is why so many pilot budgets get signed off by someone who never sees the review cost line.
The market data suggests the saving is real only where volume is high. Aitana's reported earnings of up to €10,000 a month against just over €1,000 per advert describe a creator who books many adverts, not a brand that saved on one campaign.

Before a single post goes out, the program needs four checks, and each one has an owner.
Two Topkee notes on adjacent ground are worth reading before the pilot brief is signed. Emails that answer: AI automation for business data shows how automated replies get reviewed before they reach a customer, and Adobe Firefly as a creative studio: what to automate covers which parts of a creative workflow are worth handing over at all.
A pilot that fails any one of these checks should not launch. A pilot that passes all four should still start narrow, on one channel and one audience segment, with the stop threshold in writing.
Sentiment is the metric most likely to mislead a marketing team, so it deserves its own caution.
In Topkee's vocabulary, sentiment refers to the tendency of the wording in a text. It is not customer satisfaction, and it is not trust. A positive emoji count says something about how people phrased their reaction; it is a proxy, and the moment a report calls it a trust score, the number starts misleading the room.
A workable measurement stack has three layers, and it is the sort of reporting any ai automation for business deployment should hand over documented. Interaction metrics, such as impressions, clicks, engagement, followers and enquiries, show reach and response. Sentiment read from emoji and word use shows tone in comments. A human-creator benchmark, run on the same channel at the same time, shows whether the AI persona performed differently from a real creator rather than differently from nothing.
Teams that buy ai automation for business support usually get this layering written into the handover, because a dashboard without a benchmark answers nothing. Without it, a novelty spike reads as success, and the drop that follows reads as a mystery.
Searches such as Unisoft ai 課程 and AI agent classes dominate this topic, and they answer a different question: how to learn the tools. They leave open the question a brand has to decide, which is whether an AI persona belongs in a campaign and how its performance gets measured. The framework above fills that space, and the four checks give a reader who came in through training material a way to apply it to a real budget.
The interesting question for a brand is not whether a persona can be produced but whether the system behind it can be audited afterwards. That is where Topkee's documented work sits: not as an influencer product, and never described as one, but as the surrounding automation and measurement layer.
Topkee's AI Agent development covers AI workflow automation, application development and CRM and ERP system integration, with the stated design rule that data must have a source, execution must have permissions and results must be verifiable. In the workflow automation design, triggers, inputs, outputs and human confirmation points are specified up front, and automatic publishing outside the agreed scope is explicitly not done by the Agent.
The measurement side is where a sentiment pilot gets its footing. Topkee's social media analysis reads sentiment from users' use of emoji and word choice, and tracks impressions, clicks, engagement, followers and enquiries across channels in one dashboard. It is a modest behavioural proxy, and the honest way to present it is beside interaction metrics rather than instead of them.
For a Singapore team, the same discipline applies to erp integration services as to a campaign system: check interface availability, authorisation and data quality before the first run, and keep a named person who confirms what gets published. Where a brand already runs erp integration services across its channels, the campaign data often lands in the same place as the sales data, which makes the benchmark easier to hold than a standalone social report would.

Three things are worth tracking into the next quarter.
The first is whether disclosure becomes a creative convention rather than a warning label. If platforms standardise the AI badge, the audience's tolerance typically rises, because the surprise disappears. The second is whether the trust figures move as more brands run labelled campaigns and the category matures. A team without its own data will keep asking whether ai automation for business changes the answer, when the answer depends on its own audience. The third item is your own benchmark data, which is often what a team first buys ai automation for business work to get: after one pilot, you will know how your audience reacts, which is worth more than any survey of a generic consumer.
The Clutch survey suggests the damage concentrates on deceptive use rather than on AI itself. 64% of consumers would trust a brand or creator less if AI were used to misrepresent products, while 42% view AI positively when it is used for educational content.
According to the ResearchGate-published study, 26.4% of consumers would actively engage with an ad featuring an AI influencer. 45.3% would view it without taking action and 28.3% would ignore it completely.
Consumers expect it: nearly 90% say disclosure of AI-generated content is important. Some markets now require it, including New York for certain ads featuring AI-generated human performers as of June 2026, and EU transparency obligations since 2 August, 2026.
No. Sentiment read from emoji and word choice is a proxy for how people phrased their reaction. It needs an interaction baseline and a human-creator benchmark before it can support a decision.
Only if the category does not depend on lived experience and the team can staff the four checks, from audience fit to a stop threshold. Otherwise a labelled human creator remains the lower-risk route.