Founders still ask me the question they asked in 2019: what does a good backlink cost. This September I finally answered it properly. ESBO Ltd, the link building and digital PR agency I run, exported its entire publisher database, 16,625 sites across 53 languages with verified authority scores and traffic, and published the numbers as the State of Link Building and Brand Mentions 2026. The median sponsored article costs US$570. That turned out to be the least interesting thing in the data.

The publishers who charge most sell no equity

One publisher in ten sells nofollow links only, meaning links that search engines are explicitly told not to count. Move up the authority ladder and the share climbs: at a Domain Rating of 80 and above, the top tier of site authority, 39 per cent of publishers are nofollow only, and among sites with more than a million monthly visits, 36 per cent. Those publishers quote a median of $1,840 per article. Everyone else quotes US$510.

I expected the opposite when I ran the numbers. Instead the pattern is clean: 3.6 times the price for placements that pass none of the search equity buyers think they are paying for. Large publishers do not sell link equity. They sell their audience and their name, and they price that as advertising, with the hyperlink as a formality.

Also Read: The founder-to-minister pivot isn’t the problem, ASEAN’s missing governance infrastructure is

The machines already agree with them

Two public data sets explain why that pricing now makes sense. Ahrefs tested which factors correlate with a brand’s visibility in Google’s AI Overviews across 75,000 brands, and branded web mentions correlated at 0.664 against 0.218 for backlinks. On that measure, being talked about predicts AI visibility about three times better than being linked to. Muck Rack, analysing more than 25 million links cited by ChatGPT, Claude and Gemini, found earned media accounts for 84 per cent of AI citations, and paid or advertorial content for 0.3 per cent.

Put those together and a sponsored placement in 2026 buys three separate things: an audience that reads it, a brand mention in a context machines index, and sometimes a link. The market has repriced from the top down, and it is the mechanism under everything I have written in this column since February about becoming the source machines quote.

The Southeast Asia discount, and the trap inside it

Now the part that matters for this region. English is not one market in the data. English-language sites whose readers sit mainly in Western countries carry a median price of US$593. The 1,697 English sites whose readers are mainly in South and Southeast Asia or Africa: US$150. Sites read mostly from India: US$100.

Read one way, that is a genuine buying opportunity. A startup selling into this region can appear in front of its actual buyers for a quarter of Western prices, and the same gap runs through local languages, where Central and Eastern European placements cost half of what Western European ones do.

Also Read: Taiwan bets on Gen Z founders to move beyond its chip-supplier image

Read the other way, the cheap end is where the trap sits. Disclosure collapses as prices fall. Estonian publishers state a sponsored label 83 per cent of the time, Indonesian publishers 9 per cent. And if machines cite paid content 0.3 per cent of the time, a US$60 undisclosed link on a site nobody reads buys neither search equity nor machine memory. It buys a line in a report.

Spending a small budget like it is 2026

Four adjustments follow for a founder with modest money.

Buy the audience and the mention, not the metric. A US$300 placement whose readers are your actual buyers beats a US$900 one chosen for its authority score.

Treat the link attribute as a bonus. If the article is worth publishing with a nofollow link, it is worth publishing. If it only makes sense followed, you are buying the wrong thing.

Move the saved money to earned coverage. Journalists, reviewers and industry newsletters generate the 84 per cent, they cost effort rather than invoices, and publishing original numbers about your market is still the fastest way to interest them.

Measure mentions. Most teams still count referring domains. Start counting how often your name appears, per market, in contexts machines read, because that is the number moving your AI answers.

The price lists are telling founders something the industry took years to admit. The most sophisticated publishers quietly stopped selling links some time ago. They sell being known, and that is the part the machines keep.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post The most expensive links aren’t really links: What 16,625 publisher price lists tell SEA startups appeared first on e27.

Most AI writing about work still frames the shift too narrowly. The conversation often starts with a familiar question: which tasks can AI do? That is a useful starting point, but it misses the more important operational change.

The deeper shift is not that AI can write, summarise, classify, or draft faster than before. It is that AI is starting to sit inside the flow of work itself. It can monitor inputs, prepare actions, surface missing information, follow rules, ask for approval, and help move a recurring job forward. That does not make it an employee in the literal sense. But it does change what humans need to manage.

This matters because most work does not break at the level of one isolated task. It breaks when the next step is unclear, the wrong person owns the follow-up, the context is scattered across tools, exceptions are missed, or a decision goes out before it has been checked. In other words, work often fails in the operating layer between tasks.

That is where AI is becoming more consequential.

A support team does not only need a draft reply. It needs the issue summarised, the account context pulled together, the relevant policy found, the refund rule checked, the risky cases flagged, and the final action paused for review when money or commitments are involved. A sales team does not only need notes cleaned up. It needs lead context structured, missing information identified, the next step drafted, the CRM updated, and stale leads surfaced before they disappear. A finance team does not only need extraction. It needs exceptions separated from routine items, approvals routed correctly, and anything customer-facing or irreversible held back until someone signs off.

These are not just tasks. They are managed responsibility systems.

That is the more useful way to understand the shift. AI is not only helping people complete individual actions. It is helping organisations redesign how recurring responsibilities are carried, checked, and escalated.

This is also why the “AI employee” framing can mislead. An employee is not just a bundle of outputs. An employee sits inside accountability, authority, escalation paths, and consequences. Most organisations are nowhere near handing all of that over. What they are doing instead is more specific and more realistic: they are letting AI prepare, route, monitor, and sometimes execute bounded steps inside a workflow, while humans remain accountable for judgment and exceptions.

That distinction matters.

Also Read: AI won’t just replace jobs. It will redesign how companies work

When companies treat AI like a smart drafting box, the human still carries nearly all of the responsibility. The person has to remember what to do, gather the context, issue the prompt, inspect the output, send the result, and remember to follow up later. The AI helps with one segment of the work, but the burden of orchestration remains largely human.

Once AI is connected to tools, triggers, records, and approval points, the shape changes. The system can notice a new request. It can compile the relevant context. It can draft the likely next action. It can flag missing data. It can ask for approval before a financial, legal, or customer-facing step is taken. It can record what happened and surface what still needs attention.

The responsibility does not disappear. It gets redistributed.

That redistribution is where many teams are still underestimating the management challenge. As AI moves closer to action, organisations need clearer decisions about where approval is required, which actions are reversible, what should always stay human-led, and what evidence the system should attach before asking for sign-off. A bad draft is one problem. A bad action wrapped in a polished draft is another.

This is why operational design matters more than prompt cleverness. The hard questions are not only about what model to use. They are about what should trigger the workflow, what context should be assembled automatically, what counts as a routine case, what should pause for review, where the result should be stored, and what should happen when the system encounters ambiguity.

Also Read: Your startup has an AI strategy. Does it have a human strategy?

These questions sound mundane compared with product demos. They are also much closer to how work actually succeeds or fails.

That has strategic implications for hiring and management too. As more recurring work is repackaged into supervised AI systems, strong individual contributors will need to think more like operators. They will need to define decision points, identify failure states, design escalation paths, and describe what good output actually looks like in context. The value shifts away from doing every small step manually and toward designing how those steps should move.

This does not mean every team should rush into deep automation. Some workflows are too messy. Some decisions are too sensitive. Some domains generate too much downstream risk if the system acts too early. In many cases, the best design is not full autonomy but staged assistance: prepare the context, draft the recommendation, require approval for the critical move, and keep a visible record.

That may sound less dramatic than the idea of AI becoming an employee. But it is probably closer to what serious adoption will look like inside real organisations.

The next phase of workplace AI is not just about replacing effort. It is about redesigning responsibility.

That is the real management shift. The question is no longer only which task AI can perform. It is which recurring responsibility can be turned into a supervised system with clearer triggers, better context, tighter review, and fewer dropped handoffs.

That is where AI starts changing work more deeply.

—

The post AI will not become your employee. It will change what work has to be managed appeared first on e27.

AI technology is transforming the agricultural industry, changing how the outside world views farming and creating a data-driven system that ensures precision and maximises yields.

Imagine a world where technology tells you remotely how many nutrients are lacking and how plants are showing signs of stress and so on. It’s a beautiful experience, not science fiction like many thought. It’s a reality being practised by some big farms around the globe.

Ways in which AI is transforming agriculture

• AI makes caring for every plant and animal less laborious

For centuries, farmers have relied on just walking around the farm field to take records of animal and plant performance. Oftentimes, it is extremely challenging to manually care for every single plant and animal, especially for commercial farmers. Data and images of farm activities are seamlessly fed into AI systems via drones, satellites, and ground sensors. Making it much easier for farmers to remotely spot problems that the human eye could struggle with. For example, early disease detection, nutrient shortages and water stress control. In fact, it’s like having a health app that swiftly informs you before trouble spreads. 

• AI helps farmers to know when and how much to feed and water plants and animals

As important as water, feed, supplements, and fertilizers are, excessive application can cause other problems. AI-powered soil sensors, weather data, and smart irrigation and feeding systems can help farmers figure out how much each part of the field needs (either plant or animal) and when they need it. Typically, this helps farmers to save money by controlling waste of resources. 

• AI helps farmers to solve the weather and planting season puzzle

Accurate weather prediction has been a long-standing issue for farmers. AI can turn farmers’ “gut feeling” into accurate predictions on when to plant, what kind of crop to plant, reveal dry and rainy seasons and lot more. Over the years, making predictions has been a struggling art for farmers, but AI makes it seamless as it digests years of weather records, market trends, and soil data to give farmers a clearer recommendation. Of course, this technology is not at its perfect state yet, but it’s a huge step-up in agriculture.

• AI takes away the back-breaking tasks from farmers’ shoulders

For decades, farmers have been relying on human labour for tasks like weeding, planting and harvesting. Typically, these tasks are super stressful, back-breaking, and even expensive. However, well-designed technological farm tools like robotic weeders, autonomous tractors, smart harvesters, and so on that are guided by artificial intelligence (AI) fill the gap. They work longer hours without getting tired. Large farms that are short on labour can leverage it for their seasonal operations.

Also Read: Why Southeast Asian agritech must build for acquisitions, not IPOs

Core challenges to look out for

The promise of this technology is very real and enticing: better yields, much less waste, making smarter decisions and a lot more, but the everyday realities of farmers make it uncertain and a lot more challenging for many farmers to align with it. Here are major relatable hurdles to look out for;

• High initial investment cost and uncertain ROI

According to Mckinsey’s Global Farmers Insight in 2024, one of the major barriers to agricultural technology is high cost. This research revealed that European and North American countries are leading in global agricultural technology adoption. Meanwhile, about 52% of North American and 48% of European Farmers cited “huge costs” as the biggest challenge of adopting agtech; and about 40% of North Americans also reported that “unclear ROI” stands as a huge barrier to adoption.

Compared to large agribusiness farmers operating in millions of hectares, small and mid-size farms feel this the most. It’s very difficult for small or mid-sized farms to spend thousands of dollars on drones, sensors, software subscriptions, and a lot more with uncertain ROI. In fact, most of these farmers need clearer proof that their investments would pay off before investing in any seasonal budget. 

• Displacement of human labourers

In rural and regional communities where farms have adopted AI automation, the displacement of human labour will be high because the machine can run human operations for hours without getting tired, and as such, there won’t be a need for extra labour to attract extra cost. However, there is a growing need to have skilled personnel to operate those machines excellently.

• Network, power and connectivity barrier in rural areas

Here is another crucial barrier to look out for. From all indications, almost every AI-powered tool needs a stable network supply to enhance seamless data communication. And all devices need electricity to operate efficiently.

This is a roadblock for rural and regional farmers where the network is completely unreliable. Without a stable network, soil sensors can’t communicate to the cloud, apps wouldn’t be able to pull weather models, and even cameras won’t be able to spot pests in real time.

• Operational knowledge complexity

Farmers who are not familiar with sophisticated devices would find it daunting to operate AI tools. The language barrier ( to read through the manual), the huge numbers of low literacy within those regions and the limited number of training they might receive make it very challenging to adopt.

• Data privacy and trust struggle

Here is another barrier you can’t ignore, as the success of AI tools depends on their ability to learn from large numbers of datasets accumulated from the farm. But the big worry is always where this data is stored, how accessible the data is to farmers and a lot more. Meanwhile, there are farmers who are intimately accustomed to their farms such that they feel it’s unsafe to share sensitive data with AI. Imagine a device telling you when to irrigate, feed or apply fertiliser without explaining why; this makes some farmers feel like they are handing control to strangers.

Also Read: Agritech’s next business model may not charge the farmer

AI advancement in technology and how it’s helping farmers today

According to futurist Jim Carroll, AI advancement in agriculture offers many promising pathways in both crop and animal production. It’s already delivering exciting benefits like boosting yields, reducing waste, and improving animal welfare. Moreover, fascinating agricultural technology companies are tirelessly working to improve farming across the globe. Here are a few;

• Inventions towards targeted weed control/precision spraying with strong global recognition are John Deere See & Spray and Carbon Robotics LaserWeeder: They are advanced computer-vision and machine learning precision agricultural AI systems that can swiftly identify the target(weed) in real time and spray only them, not the entire field
• Invention towards crop monitoring, disease and pest detection: Taranis and Plantix are high-tech inventions that are AI-powered for early detection of pests and diseases, nutrient deficiencies and more.
• Invention towards advisory chatbots and smallholder tools: Farmer.Chat and Kisan e-Mitra are AI chatbots that help farmers access information on schemes, weather, pest and disease management, and more. They are often used in Africa.

In conclusion, AI is already being used on real farms, and it’s transforming farm activities from constant worries to something smart. While its primary goal is to help farmers grow more food with less waste, fewer chemicals, and almost no guesswork, it doesn’t mean farming suddenly becomes easy.

—

The post AI vs agtech: How AI is revolutionising agriculture appeared first on e27.

A few years ago, the early stages of a video project often began almost from zero. We would talk through the client’s direction, work out a script, develop the visual approach and gradually turn a loose idea into something the client could react to.

That part has changed. Today, a client can give me a rough direction and I can use AI to produce a first script quickly, revise it and put something concrete back in front of them. From there, we can move into the rest of production with much less work spent creating everything from scratch.

The first draft became easier. The decision often did not.

A client may know what the video is for and still have difficulty describing what it should look or feel like.

Usually I start by asking whether there is a reference video or image close to what they have in mind. If there is, we can talk about something specific. If there is not, I will often find several examples myself and ask what feels right and what does not.

The conversation often becomes clearer at that point. One direction may feel too polished. Another looks too much like an advertisement. The pacing may work while the characters do not. Sometimes ruling something out tells us as much as choosing something.

Once the direction becomes clearer, the conversation also becomes more specific. We can move from “something like this” to decisions about the script, visual style, characters, scenes and eventually the storyboard. Each step narrows what the production side has to interpret on its own.

This is why I do not think of requirement discovery as a single conversation at the beginning of a project. In practice, it develops through a sequence of increasingly concrete choices. The earlier those choices become clear, the less likely the production team is to spend time building in the wrong direction.

Part of my job, then, is not simply to take an instruction and turn it into a video. It is to help the client arrive at the instruction in the first place.

AI helps me get something concrete in front of the client earlier. It does not remove the conversation that follows.

A different kind of uncertainty can appear later.

Also Read: SEA startup funding jumps to US$7.25B, but most founders are still waiting

I have worked on projects where the script was confirmed, the main visual direction was confirmed and the work had already reached a rough cut. From the production side, the major choices seemed settled. Then another person or department inside the client organisation reviewed the material, and earlier decisions reopened.

What looked like approval from one side of the project turned out to be one stage in a longer internal process. Only then do we discover that the decision was not actually closed.

Making another version may be straightforward. Knowing whether the people reviewing it are the people whose decision will hold can be harder.

A late change is not just another prompt or another edit. By that point, several parts of the work may already depend on the earlier decision. A changed script can affect images; changed images can affect video generation and editing; feedback then has to travel back through the people doing each part of the work.

Faster generation makes the replacement work easier. But the change still has to be passed back through the people doing the rest of the work. In some projects, that coordination takes more of my attention than producing the next asset itself.

This has changed where I spend my own time.

Once a requirement is clear, more of the execution can be handed to other people. Storyboarding, image generation, video generation and editing can all be handed over. I can set out how the work should be done and check it as it progresses.

That still leaves a lot of work with me: understanding what the client means, collecting feedback, passing that feedback back to the production side and checking that the next version still matches the agreed direction.

Also Read: SEA startup funding jumps to US$7.25B, but most founders are still waiting

I can hand off more of the making itself. The client conversations still come back to me.

A new version can be useful because it changes that conversation. The client becomes clearer about the direction. Two people who appeared to agree react differently to the same material. Someone new joins the review and reveals that an earlier approval was not final.

At other times, the pictures change, the script changes and the edit improves, but the same unresolved question remains behind the work.

That is the difference I now pay attention to. When the next version becomes easier to produce, I ask what changed because we made it. If the client is no clearer about what should be made and someone else still needs to approve it, we have produced faster without getting much closer to done.

We may simply have reached the same unresolved decision sooner.

—

The post AI made the first draft easier. It didn’t make client decisions easier appeared first on e27.

For many people, foot pain is ignored until it starts changing everyday behaviour. They walk less, avoid stairs, switch shoes, or live with knee and back discomfort that seems unrelated. GoodARCH is betting that a five-minute scan can narrow the gap between early signs and medical attention.

The Asian arch support brand, operated by Taiwan-based Homeway Technology, has invested nearly US$230,000 to develop an artificial intelligence-powered foot mapping system, which it is now introducing in Malaysia. The system generates a personalised foot assessment in about five minutes, using image recognition and footprint mapping to help users understand their arch structure and spot imbalances earlier.

Also Read: From smart rings to health coaching: AI and the new preventive healthcare paradigm

The technology is available at several locations, including GoodARCH’s headquarters on Jalan Ampang in Kuala Lumpur, as well as in Penang, Batu Pahat and Johor Bahru. The company said it also plans to work with local health management providers as it expands in the country.

The launch sits at the intersection of two trends that are becoming more visible in Southeast Asia: consumers taking a greater interest in preventive health, and wellness brands using AI to turn quick assessments into personalised recommendations. In markets such as Malaysia, where private healthcare costs are a concern and an ageing population is putting more pressure on the health system, tools that help people identify potential issues earlier are drawing growing commercial attention.

From arch support to AI assessment

GoodARCH was founded in 2002 by Dr Hsieh Chin-Hsing. It began with arch support products before expanding into health footwear and everyday wellness support. Its latest AI foot mapping system was developed over 12 months by an 11-member team spanning medical engineering, business and edge computing.

According to the company, the system builds on its earlier use of cloud-based image recognition and reusable silicone footprint mapping. In practice, the pitch is straightforward: users receive a quick assessment of their foot structure, which can then inform the choice of arch support or related products.

GoodARCH said its technology has so far supported foot structure assessments for more than 300,000 users. Each assessment is paired with its Far-Infrared Arch Support insole, which the company says is designed for stability and shock absorption.

“GoodARCH has continued to invest in foot health technology and research, evolving from infrared-based arch support solutions to graphene technology, proprietary Torsion Field Energy technology, and now AI-powered digital foot mapping,” said GoodARCH Chairman Hsieh Ming-Chia.

The company has also extended its materials and wellness technologies into Health Rhythm, a physiotherapeutic recliner designed to support circulation and sleep through a passive routine. Its core technologies have received medical device approvals and National Quality Award certification in Taiwan, and it holds patents in markets including Malaysia, mainland China, Hong Kong, South Korea, the Philippines, Thailand and Indonesia.

Why Malaysia matters

Malaysia is a logical testbed for GoodARCH’s next phase. The country has a relatively developed private healthcare and wellness market, urban consumers familiar with mall-based health screenings, and a growing middle class willing to spend on products that promise comfort, mobility and long-term wellbeing.

Also Read: The US$500 fix that could unlock a lifetime: How MiracleFeet is closing Asia’s clubfoot gap

There is also a real public health backdrop. GoodARCH cited data suggesting that up to 75 per cent of people will experience a foot problem during their lifetime. Separately, a study of 190 students by the International Islamic University Malaysia found that 26.3 per cent had flat feet.

Flat feet do not always cause pain, and not every case requires treatment. But arch structure can influence balance, gait and load distribution across the body. Over time, persistent misalignment may contribute to discomfort in the ankles, knees, hips or lower back, particularly among people who stand for long periods, wear unsuitable footwear, or have age-related joint problems.

Malaysia’s ageing population makes this more relevant. Data from the Ministry of Health Malaysia and the Malaysian Orthopaedic Association show that 30 to 40 per cent of Malaysians aged 60 and above suffer from knee osteoarthritis. While osteoarthritis has multiple causes, including age, weight, genetics and injury, long-term biomechanical stress can be a contributing factor.

For GoodARCH, the opportunity is not to replace clinical diagnosis but to make the first step less intimidating. “Many people only start paying attention to their feet when pain or difficulty walking begins to affect daily life. AI gives us an opportunity to change that by making foot assessment a simpler first step toward greater awareness and earlier action,” said founder Dr Hsieh Chin-Hsing.

That distinction matters. AI-assisted consumer assessments can be useful for screening and education, but they also risk overpromising if positioned as medical diagnosis. The success of such systems depends not only on speed and convenience, but on clear communication about what a scan can and cannot tell a user, especially when the assessment leads straight to a product recommendation from the same company.

A competitive space for personalised foot care

GoodARCH is entering a market that already includes global and regional players using scanning, pressure mapping and customisation to sell insoles and footwear. Aetrex offers foot scanning technology through retail partners, while FootBalance provides customised insoles shaped around in-store analysis. Dr. Scholl’s has long used kiosk-based foot mapping in mass retail, and Superfeet has built a strong brand around performance and comfort insoles.

In Southeast Asia, competition also comes from podiatry clinics, physiotherapy centres, orthopaedic footwear providersand sports retailers offering gait analysis. This makes GoodARCH’s challenge twofold: it must convince consumers that its assessment is credible, while showing partners that its system can fit into existing wellness, rehabilitation or retail workflows.

Also Read: Why Southeast Asia’s next healthtech winners will be built around healthcare workflows, not just AI

The broader direction, however, is clear. Healthcare is moving beyond hospitals and clinics into pharmacies, gyms, shopping centres and homes. For healthtech startups in Southeast Asia, this shift opens space for tools that are faster, cheaper and easier to access than traditional specialist appointments.

GoodARCH’s Malaysian rollout is therefore less about a single foot scan than a wider question: how much preventive healthcare can be delivered before a person becomes a patient? If the company can answer that with enough clinical discipline and consumer trust, foot mapping may become part of a much larger wellness stack.

The post GoodARCH launches AI foot mapping in Malaysia with US$230K healthtech investment appeared first on e27.