Pattern 1: Facts in plain HTML beat facts in scripts
If a price, address or product name only appears after JavaScript runs, many AI crawlers never see it.
Most modern websites load a skeleton page first and then fill in details with JavaScript, prices from a pricing API, store locations from a map widget, reviews from a third-party embed. A person waits half a second and sees everything. Many AI crawlers don't wait. They fetch the initial HTML, extract the text, and move on. Whatever arrived later simply isn't there.
The result is a page the AI treats as incomplete: a product with no price, a clinic with no address, a service with no package details. When the assistant builds an answer to “how much does X cost in Malaysia?”, it quotes a competitor whose price was sitting in the HTML.
How to check in five minutes
- Open one of your key pages in a browser.
- Right-click and choose View Page Source (not Inspect, Inspect shows the page after scripts run).
- Press Ctrl/Cmd + F and search for the price, address or product name.
- If it's there as plain text, you're fine. If it isn't, it's being rendered by JavaScript.
- Repeat on 10 to 20 of your most important pages.
Structured data (schema.org Product, Offer, LocalBusiness) helps, but it's a supporting signal, not a replacement. Put the fact in the visible HTML first, then mark it up.
What to do in the next 30 days
Run the view-source check on your top 20 pages. Anywhere a key fact is missing from the raw HTML, raise a ticket to render it server-side. On Next.js, Shopify Hydrogen or other headless stacks this is usually a small change at the rendering layer.
Pattern 2: Repeat coverage on trusted publishers beats new domains
A second or third mention on a publisher AI already cites is worth more than a first mention on a site it has never heard of.
AI assistants don't read the whole web every time you ask a question. They lean on a relatively small pool of sources that have proved reliable, established news sites, well-known trade publications, reference sites. When one of those sources describes your brand consistently across several articles, the assistant becomes confident about who you are and what you do.
Most link-building programmes are built to maximise new referring domains. That's a reasonable goal for classic SEO, but it spreads effort across sites AI systems may never consult. For AI visibility, depth on a few trusted publishers usually wins.
Signs a publisher is already in the AI source pool:
- It appears as a cited source when you ask ChatGPT, Perplexity or Google's AI Mode broad questions about your industry.
- It ranks on page one for informational queries in your category.
- Other publishers quote or link to it as a reference.
What to do in the next 30 days
List every publication that has mentioned your brand in the last two years. Ask three AI assistants ten category questions and note which of those publications they cite. Pitch the top five a fresh angle this quarter, new data, a founder view, a customer story.
Pattern 3: Honest comparison pages punch above their weight
People ask assistants “X or Y?” questions. Pages that answer that exact question get cited.
“Should I use Xero or Bukku?” “Is a press release or a guest post better for a product launch?” These are structured questions, and assistants look for structured answers. A general category page sits at the wrong level; a direct comparison sits at exactly the right one.
The worry is usually thin content. Search engines do penalise mass-produced comparison pages, thousands of auto-generated “A vs B” pages with no real testing behind them. A handful of carefully researched comparisons is a completely different thing.
What makes a comparison page citable:
- A clear verdict for each use case (“Choose A if…, choose B if…”), not a fence-sit.
- A side-by-side table with real prices, limits and features.
- Honesty about where your own product loses.
- A visible last-updated date.
What to do in the next 30 days
Pick your three most important commercial topics. For each, write one comparison against the alternative customers genuinely weigh you against, roughly 1,500 to 2,500 words with a scoring table and a clear verdict. Link to it from your service page and your most-read articles.
Pattern 4: Third-party signals beat domain rating
AI systems confirm a brand is real by checking what others say about it. Reviews and independent mentions are the strongest evidence.
Domain rating (DR) and domain authority (DA) measure backlinks. They say little about whether real customers exist or what they think. Review platforms and independent coverage answer exactly those questions, and assistants read them, including the language inside reviews, which tells them what you're known for.
In Southeast Asia, the platforms that matter depend on the category: Google Business Profile reviews almost everywhere; Facebook recommendations for many local services; Shopee and Lazada ratings for consumer products; G2, Capterra and Trustpilot for software and B2B services.
Never buy or gate reviews. Platforms remove them, and suspicious review patterns hurt trust with both people and machines. A steady trickle of genuine reviews beats a sudden burst.
What to do in the next 30 days
Identify the two or three review platforms that appear when you ask an assistant about your category. Add a review request to your post-purchase or post-project email. Aim for a steady flow of genuine reviews every month rather than a one-off push.
Pattern 5: Fresh and focused beats long and stale
A 1,200-word page updated this quarter often outperforms a 4,000-word page nobody has touched in two years.
The old advice was to write the longest, most comprehensive page on a topic. For fast-moving subjects, regulations, pricing, platforms, AI itself, freshness now matters more. Assistants prefer current information, and readers engage more with pages that reflect what's true today.
A refresh doesn't mean a rewrite. Replace outdated figures, add one new example, explain one new development, remove anything no longer true, and update the date only when the content has genuinely changed.
Rough refresh cadence:
- Fast-moving topics (finance, regulation, software, AI): every 90 days.
- Commercial pages you depend on for leads: at least every 12 months.
- Evergreen explainers: review yearly, update when something changes.
What to do in the next 30 days
Pull your ten highest-value pages and check their last-updated dates. Refresh anything older than a year, starting with the pages that bring in the most enquiries.
Pattern 6: Depth on one topic beats volume on one keyword
A site that covers a subject from thirty genuine angles looks like an authority. A site with one optimised page looks like a guess.
When an assistant decides which brand to name as the expert on a subject, breadth of coverage on that subject is a strong clue. A cluster of interlinked pages, the main guide, the how-tos, the comparisons, the FAQs, the local variations, shows sustained expertise.
The trap is producing thirty thin pages to fake depth. Templated, low-effort content is exactly what recent search updates target. Every page in the cluster has to answer a real question that the others don't.
Where to find genuine angles:
- Search Console queries you already get impressions for.
- “People also ask” boxes on Google.
- Questions your sales and support teams answer every week.
- Prompts you try in ChatGPT and Perplexity that return weak or outdated answers.
What to do in the next 30 days
Choose one topic you want to own. Map the 30 to 50 questions people ask about it. Publish or refresh five to ten pages this quarter, all linking to one central guide.
Pattern 7: In Asia, one brand needs one identity in every language
If English and Chinese sources name you differently, AI systems may treat you as two unrelated companies.
Brands in Malaysia, Singapore, Hong Kong and Taiwan often appear in English and Chinese media, sometimes with a transliterated name, sometimes with a translated one, sometimes with both. When different outlets use different versions, it's harder for an AI system to connect the coverage to the same company.
- Choose one official Chinese name (and romanisation if relevant) and use it everywhere.
- Put both names together on your About page, e.g. “Sojuva (索久瓦新闻社)”, whichever version you choose.
- Include both names in press-release boilerplate so every outlet copies the same pairing.
- Keep company registration details, address and founder names identical across languages.
What to do in the next 30 days
Search your brand in English and in Chinese. Note every name variation you find. Pick the official pairing, update your own site and boilerplate, and ask the publishers who covered you most recently to correct any mismatches.
What to do this quarter
- 1Put key facts in plain HTMLRun the view-source check on your top 20 pages and fix anything rendered only by JavaScript.
- 2Map your trusted publishersFind which outlets that already covered you are cited by AI assistants, and pitch the top five again.
- 3Publish three honest comparisonsOne per core commercial topic, with a table and a clear verdict.
- 4Systemise genuine reviewsAsk every customer, on the platform that matters in your category.
- 5Refresh your top ten pagesCurrent data, one new example, remove what's outdated.
- 6Build one topic clusterFive to ten genuinely distinct pages around one central guide.
- 7Lock your bilingual identityOne English name and one Chinese name, used identically everywhere.
Frequently asked questions
Which AI assistants should we focus on first?
For most businesses in Southeast Asia: ChatGPT (largest user base), Google's AI Overviews and AI Mode (because they sit on top of Google Search), then Perplexity (which cites sources very visibly). Track Gemini and Copilot too, but the first three cover most of the answers your customers see today.
How do I know if our prices are rendered by JavaScript?
Open a product or pricing page, choose View Page Source, and search for the price. If it's in the raw HTML, it's server-rendered. If you can only see it on the live page, it's added by JavaScript and many AI crawlers will miss it. Check 10 to 20 pages, not just one.
Is schema markup still worth doing?
Yes, as a supporting signal. It helps machines interpret facts that are already visible on the page. It doesn't rescue a fact that isn't in the HTML at all, so fix visibility first and add schema second.
Will comparison pages get us penalised?
Not if they're genuine. Search engines target mass-produced comparison pages with no real research. Three to five carefully researched comparisons per core service, each with a clear verdict and honest limitations, are normal, useful content.
Do press releases help with AI visibility?
They can, when the release is newsworthy and lands on publishers that AI systems already trust. A release syndicated only to low-quality sites adds little. Consistent facts, names and figures across every placement matter as much as the number of placements.
How long before we see changes in AI answers?
Technical fixes like HTML visibility can show up within weeks, once pages are re-crawled. Reputation signals, coverage, reviews, topical depth, build over months. Check the same set of prompts monthly so you can see the trend rather than one-off results.
Related
More articles