Last week, a friend showed me his AI-generated keyword list for a nasi lemak delivery business in Penang. The tool suggested he target "nasi lemak delivery" with a reported search volume of 12,000 monthly searches.
Sounds great. Except the number was completely made up.
The tool had hallucinated the data. It looked real. It felt real. It sat neatly in a spreadsheet alongside other keywords with equally confident-looking numbers. But when we cross-referenced against actual search data, the real volume was a fraction of that figure.
This is the world of AI SEO tools in 2026. Genuinely powerful, occasionally brilliant, and sometimes dangerously wrong. The trick is knowing which is which.
What AI SEO tools actually do well
Let's give credit where it's due. AI has made certain SEO tasks dramatically faster and, in some cases, better.
Content briefs are a perfect example. Tools like Surfer SEO and Frase can analyse the top-ranking pages for a keyword, identify common subtopics, and generate a structured brief in minutes. What used to take an SEO specialist an hour now takes seconds. The output isn't perfect, but it's a strong starting point.
AI-generated meta descriptions are another win. Writing 200 unique meta descriptions for an e-commerce site used to be mind-numbing work. Now you can generate solid first drafts in bulk. They still need a human eye — more on that shortly — but the time savings are real.
Here's a quick breakdown of where AI shines versus where it struggles:
| Task | AI Performance | Human Review Needed? |
|---|---|---|
| Content brief generation | Excellent | Light editing |
| Meta description drafts | Very good | Moderate review |
| Keyword clustering | Good for English | Heavy review for BM |
| Search volume estimates | Unreliable | Always verify |
| Technical SEO audits | Excellent | Light review |
| Local keyword research (MY) | Poor to fair | Essential |
Keyword clustering — grouping related keywords by search intent — is another strength. AI can process thousands of keywords and sort them into logical groups faster than any human. For English-language markets with abundant data, this works remarkably well.
Technical SEO audits have also improved. AI-powered crawlers can identify broken links, duplicate content, and page speed issues with impressive accuracy. These are pattern-recognition tasks, and pattern recognition is exactly what AI excels at.
The hallucination problem nobody talks about
Now for the uncomfortable part.
AI tools can fabricate search volume data. Not occasionally. Regularly. And they do it with the same confidence they display when giving you accurate information. There's no asterisk. No warning label. Just a number in a cell that looks identical to every other number.
This happens because many AI SEO tools don't pull directly from reliable APIs like Google's own data sources. Instead, they estimate. They model. They predict. And sometimes, as research into AI reliability has consistently shown, they hallucinate — producing outputs that sound plausible but have no basis in reality.
For businesses making budget decisions based on these numbers, the consequences are real. You might invest months of content creation targeting a keyword that barely gets searched. Or worse, you might ignore a genuinely valuable keyword because the tool underestimated its volume.
The risk multiplies when you're working with lower-volume markets. Which brings us to the elephant in the room.
Why Malaysian SEO breaks most AI tools
Here's something SEO practitioners in Malaysia already know but rarely say out loud: most AI SEO tools essentially go silent on Bahasa Melayu keywords.
Try finding reliable search volume data for "servis aircond murah KL" or "kedai tayar berdekatan" in any major AI SEO platform. You'll get one of three responses: no data, suspiciously round numbers, or volumes that seem copy-pasted from a vaguely related English keyword.
The data gap is real and well-documented. As highlighted in a Search Engine Journal analysis of SEO tool limitations, keyword research tools consistently underperform for non-English and regional language queries. Malaysia's unique multilingual search behaviour — where users switch between English, BM, and even Mandarin within a single search session — makes this even worse.
So what do Malaysian SEO practitioners do? Many have started using ChatGPT and Claude as substitutes for BM keyword ideation. Ask Claude for a list of keywords a Malaysian homeowner might search when looking for a renovation contractor, and you'll get surprisingly useful suggestions.
But — and this is important — those suggestions come without data.
You get ideas. You don't get volumes. You don't get competition metrics. You don't get trend data. It's brainstorming, not research. Treating it as research is where businesses get into trouble.
This is both a clever stopgap and a genuine data-quality risk. The ideas might be excellent. The decision to invest RM5,000 in content around those ideas? That needs actual numbers behind it.
The human layer you can't skip
The pattern is clear. AI tools are excellent assistants and terrible bosses.
They can draft. They can sort. They can identify patterns at scale. But they can't do three things that matter enormously in SEO:
- Verify their own outputs. An AI tool won't tell you it just invented a search volume. You need a human (or a more reliable data source) to catch that.
- Understand local nuance. The difference between "murah" and "berbaloi" in a Malaysian consumer's mind is something no AI model has been trained to grasp properly. One signals cheap. The other signals value. Target the wrong one and you attract the wrong audience.
- Make strategic judgment calls. Should you target a keyword with lower volume but higher buying intent? Should you write in BM or English for a particular service page? These decisions require market understanding that AI simply doesn't have for the Malaysian context.
The best approach in 2026 isn't AI versus human. It's AI plus human, with clear boundaries around which tasks each handles.
Let AI generate your content briefs. Let it cluster your English keywords. Let it draft your meta descriptions and flag technical issues.
But keep a human in charge of verification, strategy, and anything involving Bahasa Melayu keyword data.
How to actually get reliable Malaysian keyword data
If AI tools struggle with Malaysian search data, what's the alternative?
You need tools built with access to real search data from the Malaysian market. Not estimates. Not models. Actual query data that reflects how Malaysians search — in all their multilingual, code-switching, wonderfully specific glory.
This is exactly why tools like SapuSEO exist. Rather than trying to be everything for every market, SapuSEO focuses specifically on Malaysian keyword research, analysis, and rank tracking. When you search for a BM keyword, you get data sourced from Malaysian search behaviour — not a guess extrapolated from a global model.
The difference matters. It's the difference between knowing that "servis aircond Selangor" gets real monthly searches and hoping that it does because an AI tool told you so.
If you've been relying on AI tools or global platforms and wondering why your Malaysian SEO results feel inconsistent, the data source is likely the problem.
You can start a 7-day trial of SapuSEO for just RM2 and see what real Malaysian search data looks like. Compare it against whatever your current AI tool is telling you.
The gap might surprise you.