Skip to content
Back to Blog
Insights

Perplexity's new models and AI visibility in Malaysia

By Bernard Leong

Co-Founder, Geo One

13 November 2026
Perplexity's new models and AI visibility in Malaysia

Short answer: when Perplexity swaps in a new model, the work that affects your visibility doesn't change. Three things still decide whether your firm turns up in an answer: whether Perplexity's crawler can fetch your pages, whether those pages state your facts plainly enough to be re-verified, and whether other credible sites describe your business the same way. A model change reshuffles the ranking among sources the engine is already able to cite. It does not make an uncitable source citable. So the sensible response to any platform announcement is to re-check the questions your buyers actually ask, rather than assume last month's result still holds.

This is written for owners and practice managers of Kuala Lumpur professional services firms with 2 to 50 staff: clinics, dental and veterinary practices, law and accounting firms, tuition centres, property agencies. Disclosure: it comes from Geo One, a Generative Engine Optimisation (GEO) and Answer Engine Optimisation (AEO) agency in Kuala Lumpur.

What actually changed on Perplexity?

The model layer under Perplexity has been rebuilt more than once. Perplexity made its in-house Sonar model the default for free users in early 2025, built on Meta's Llama 3.3 70B and tuned for web-grounded answers. Around the same time it released R1 1776, an open-source reasoning model post-trained from DeepSeek-R1. More recently, Wikipedia's entry on Perplexity records a multi-year agreement with Microsoft to use Azure and its Foundry service, which brings models from several other labs within reach of the platform.

That's the whole platform section, and it's deliberately short. The model decides how an answer is written and which of the available sources it leans on. It does not decide which sources are available, that was settled earlier, by the crawl. A Klang Valley accounting firm that cannot be fetched is invisible to every model Perplexity has ever shipped and every model it ships next. The engine keeps changing; your website sits still; your business is re-assessed from whatever is found. So the question worth your time is what gets found.

Can Perplexity fetch your pages at all?

Start here, because nothing else matters if the answer is no. Plenty of Malaysian SME sites sit behind a content delivery network with bot controls switched on by a developer who left two years ago. The owner has no idea AI crawlers are being challenged or blocked.

One Perplexity-specific detail is worth knowing before you look, because it causes most of the confusion we see. Perplexity operates more than one agent: an indexing crawler, PerplexityBot, and a separate user-triggered fetcher, Perplexity-User, which retrieves a page because someone in a live session asked about it. They behave differently, and they show up in your logs differently. The practical consequence is that you can see Perplexity traffic and still be absent from the index the engine draws on when nobody has named you, which is the case that matters commercially. Check for both agents, not one.

Three checks, in order. One, open your robots.txt file and read it line by line. Two, look at your bot management and firewall rules. Cloudflare's developer documentation sets out how its crawler controls and bot categories behave, and that's the page to read before changing anything. Three, check server logs for AI user agents over the last 30 days. If nothing has visited, a model change is irrelevant to you, because the model has nothing of yours to read.

We covered the wider version of this in our piece on what changed in GEO with Google and Cloudflare. Fix access first. It is usually the cheapest change on the list and the one with the clearest effect.

Do your pages state the facts a model needs to verify?

A web-grounded engine assembles an answer out of sentences it can find and trust. If your services page says "comprehensive solutions for your business", there's nothing to extract. If it says "we handle SST submissions and monthly bookkeeping for trading companies in Kuala Lumpur, with offices in Bangsar South", there is.

Write the facts out in plain sentences on the page itself, not only in a PDF or an image. Then mark them up. Schema.org defines the structured data types that describe an organisation, its address, its services and its opening hours in a machine-readable form. Structured data doesn't make an engine prefer you. It removes the guesswork about who you are, which is a different and more realistic goal.

Then make those facts match everywhere. Company name as registered with the Companies Commission of Malaysia (SSM), trading name, address, phone number, service list. If your Google Business Profile says one thing and your footer says another, you've handed the engine a reason to pick a source it's more sure about. Our six-check method for auditing a Google Business Profile walks through that part.

Where else does your business get described?

Perplexity shows its sources openly, which is why absence from third-party pages shows up so clearly there. A reader comparing sources beside the answer can see exactly who was quoted. We wrote about that in detail in our piece on Perplexity's Source Context Panel.

The work here is earning legitimate mentions on pages that already describe your category. For Malaysian professional firms that usually means the body you are already registered with and the local publications that cover your sector: the Malaysian Institute of Accountants for an accounting practice, the Malaysian Bar for a law firm, the Malaysian Dental Association for a dental clinic, the relevant estate agency register for a property firm. Then partner and supplier sites, and practitioner profiles. Not scaled filler, and not anything manufactured to look like coverage. One accurate listing on a body your profession actually belongs to is worth more than twenty thin profiles, because it is a page an engine can check and a human can verify.

Be realistic about what this achieves. Third-party mentions correlate with being cited by AI engines; that correlation does not establish causation. What it does do is give an engine somewhere to check you that isn't your own website.

How do you know whether any of it worked?

By asking the questions and recording the answers, repeatedly, across engines. Not by reading your traffic report, which mostly measures the engines that bother to label themselves.

Here is our own check, run in October 2025. We looked at 85 prompts across ChatGPT, Claude, DeepSeek, Gemini and Perplexity, the kind a buyer in Kuala Lumpur would realistically type. Geo One was named in 4 of them. We were named for "recommended AEO agency for small business in Kuala Lumpur". We were not named for "best generative engine optimisation agency in Kuala Lumpur".

That gap is the useful part. One prompt describes a buyer with our actual profile; the other is a broad superlative query where the engines lean on published roundups we don't appear in. Prompt-level tracking tells you which of the two you're losing. An overall score doesn't. If you run a dental practice in Mont Kiara, the same logic applies: you may be named for "paediatric dentist Mont Kiara" and invisible for "best dentist in Kuala Lumpur", and those two problems need different work.

Perplexity is the cheapest engine to run this on, because it lists its sources. When a competitor is named and you aren't, you can open the cited page and see what the engine actually used: a directory entry, a association listing, a news mention, their own service page. That tells you which of the four sections above you are failing, rather than leaving you to guess. Re-run the set after any noticeable platform change.

What to do first

In order: crawler access, then factual clarity and schema on your own pages, then consistency across SSM, Google and directories, then third-party mentions, then tracking. Doing step five first is how businesses end up with a dashboard and no change in it.

Geo One runs this work for Malaysian SMEs from RM500 to RM2,000 a month, delivered by our own AI agents that scan, fix and report. We don't guarantee placements, we don't control what any engine says, and we won't promise a timeline. We will tell you what was checked, what changed and what the answers said. If you want the longer version of the method, read our step-by-step guide to generative engine optimisation in Malaysia, or see what a month of work costs. You can also reach us by WhatsApp or direct enquiry through the contact page.

Frequently asked questions

Does a Perplexity model update affect whether my business shows up in answers?

A model change reshuffles the ranking among sources Perplexity can already cite, but it does not make an uncitable source citable. If Perplexity's crawler cannot fetch your pages, no model update will change your visibility. The three things that actually decide whether your firm appears are crawler access, factual clarity on your pages, and consistent third-party mentions describing your business.

How do I check if Perplexity can actually crawl my website?

Check three things in order: read your robots.txt file line by line, review your bot management and firewall rules, then look at server logs for AI user agents over the last 30 days. Note that Perplexity runs two separate agents: PerplexityBot for indexing and Perplexity-User for live session fetching. You need to check for both, because seeing Perplexity traffic in your logs does not confirm you are in the index the engine draws on commercially.

what changed in GEO with Google and Cloudflare

What should my services page actually say so an AI engine can quote it?

Write specific, plain-language facts directly on the page rather than vague phrases like 'comprehensive solutions for your business'. For example, stating the exact services you offer, the business types you serve, and your location gives an engine something it can extract and verify. Back this up with Schema.org structured data to describe your organisation, address, services and opening hours in a machine-readable form, which removes guesswork about who you are.

Which third-party sites should a Malaysian professional firm try to get listed on?

Focus on the registration body your profession actually belongs to: the Malaysian Institute of Accountants for accounting practices, the Malaysian Bar for law firms, the Malaysian Dental Association for dental clinics, and the relevant estate agency register for property firms. Partner and supplier sites and practitioner profiles also help. One accurate listing on a body your profession belongs to is worth more than twenty thin profiles because it is a page an engine and a human can both verify.

How do I tell whether AI optimisation work is actually making a difference?

Ask the specific questions your buyers would type across multiple AI engines and record the answers repeatedly over time. Tracking results at the individual prompt level shows you exactly where you are named and where you are not, which points to the specific work needed. Perplexity is the cheapest engine to test on because it lists its sources openly, so when a competitor is named and you are not, you can open the cited page and see what the engine used.

how Perplexity's Source Context Panel works

How much does this kind of AI visibility work cost for a small Malaysian business?

Geo One runs this work for Malaysian SMEs from RM500 to RM2,000 a month, delivered by AI agents that scan, fix and report. The service does not include guaranteed placements or promised timelines, and the agency does not control what any engine says. What you receive is a record of what was checked, what changed, and what the answers said.

Bernard Leong

Co-Founder, Geo One

Nearly 20 years across energy, capital strategy and applied AI, including large-scale operational data at BP. Founded SkillsMe and Cryptrain.

More about the team

Check Your AI Visibility

See how AI platforms currently view your business with a free scan.

Free AI Scan