Most Malaysian business owners arrive here the same way. Someone asks ChatGPT for a supplier in their industry, the answer names four companies, and none is yours. The next question is always: what do we fix first?
That has a real answer, and it is not a list of tactics. Order matters more than tactics. Rewriting fifty pages while a firewall rule quietly refuses the crawler that reads them is wasted money. It is the most common way an AI search optimisation plan fails in month one.
This guide lays out the AI search optimisation plan we run: what to fix in days 1–14, 15–30, 31–50, 51–70 and 71–90, and what to leave alone until later. It is written for the owner who approves the spend, not for a technical team. For wider context see our SEO services or the ZenWeb home page. Google has now published its own position on the topic, and the video below walks through it.
Source video: Nathan Gotch on YouTube
Quick Answer: Ninety days is long enough for a full crawl-and-refresh cycle across the engines, and short enough that nobody forgets what changed. Shorter measures noise. Longer lets five changes overlap so badly you cannot tell which worked.
The window is chosen for measurement reasons, not marketing ones. Engines re-crawl on their own schedule, so a page you rewrote on Monday may not be re-read for weeks. Check at three weeks and you are reading the old index.
It also forces a discipline most teams skip: one thing at a time. An AI search optimisation plan that changes robots.txt, schema, copy and listings in one fortnight produces a result nobody can attribute. That is the argument behind what actually changes between AI SEO and traditional SEO.
Not sure which stage your site is stuck at?
Our first fortnight is diagnosis only, before a page gets rewritten. See what an AI search agency actually does →
Quick Answer: The first fortnight is access work only. Confirm search and retrieval crawlers can reach your pages, that the pages are indexed, and that main content exists in the HTML rather than after JavaScript runs. No copywriting yet.
Google is explicit that eligibility comes first: a page must be indexed and eligible to be shown in Search with a snippet before it appears in any generative AI feature. Every later stage depends on it.
Four steps, in this order. Each has a clear pass or fail, so a non-technical owner can ask for the evidence.
Quick Answer: Weeks three and four are about identity. A model can only recommend a business it can name confidently — one consistent company name, address and description everywhere, with no competing versions of the same facts.
Malaysian SMEs lose here more often than they lose on content. The company is registered as “Sinar Klasik Enterprise” with SSM, calls itself “Sinar Klasik Interiors” on its website, and appears as “Sinar Klasik KL” on Google Business Profile. To a language model those are three weak entities, not one strong one. The work in this fortnight is deliberately boring:
Quick Answer: Across Malaysian SME sites we audit, the blockers are rarely exotic. Inconsistent business identity and thin, unquotable service pages appear far more often than crawler blocks, but crawler blocks are the fastest and cheapest thing to repair.
The table below is our intake pattern: how often each blocker appears, and how long it takes to clear.
| Blocker found at audit | Share of sites | Typical days to fix | Stage |
|---|---|---|---|
| Inconsistent business name or description | 71% | 5–10 | Days 15–30 |
| Service pages with no directly quotable answer | 64% | 20–40 | Days 31–50 |
| No third-party mentions outside own site | 52% | 30–60 | Days 51–70 |
| Key content rendered only by JavaScript | 23% | 7–21 | Days 1–14 |
| Retrieval crawler blocked at robots.txt or CDN | 14% | 1–3 | Days 1–14 |
| Money pages not indexed at all | 11% | 3–14 | Days 1–14 |
Source: ZenWeb client tracking, Malaysian SME audits, 2024–2026. Licence.
Read the last two rows carefully. They are the rarest problems and the cheapest to fix, which is exactly why a sound AI search optimisation plan starts there. Clear the one-day items before committing to the forty-day ones.
Quick Answer: This is content work, but not more content. Make each existing page answer its question in the opening lines, in plain sentences that survive being lifted out of context. Depth stays; burial goes.
Google is blunt about what does not help. Its guidance says you do not need to chunk content, write in a special style for machines, or add special markup, and that optimising for generative AI search is still SEO. It asks for unique content organised so a reader can follow it. Three edits do most of the work on a service page:
The formatting side is a discipline of its own, covered in formatting pages that LLMs can quote. Keep structured data for rich results even though Google says it is not required for AI features; where schema markup genuinely helps in AI search sets the honest expectation.
Quick Answer: Access fixes register within a fortnight. Entity and content work takes six to eight weeks to appear in answers. Off-site evidence is the slowest of all. Expect a flat first month even when the work is correct.
The table below tracks the median path across accounts that completed the full AI search optimisation plan, indexed so week 0 equals 100.
| Signal | Wk 0 | Wk 2 | Wk 4 | Wk 8 | Wk 12 |
|---|---|---|---|---|---|
| Pages eligible to be shown | 100 | 118 | 126 | 129 | 131 |
| Brand named in tracked prompts | 100 | 102 | 114 | 147 | 173 |
| Pages cited as a source | 100 | 101 | 109 | 138 | 166 |
| AI referral sessions | 100 | 100 | 106 | 129 | 158 |
Source: ZenWeb client tracking, Malaysian SME accounts, 2024–2026. Licence.
The first month of a correct plan looks identical to the first month of a plan that is failing. That is why the sequence has to be agreed in advance.
Want a baseline before week one?
Without a starting number, week twelve is just a feeling. Benchmark your AI share of voice first →
Quick Answer: Engines cross-check what a business says about itself against what other sites say. Weeks eight to ten earn real mentions off your own domain — listings, reviews, community answers and coverage a stranger wrote.
Read this stage carefully. Google warns against seeking inauthentic mentions. The goal is not volume; it is that the few places already describing your industry describe you accurately. Three areas usually pay for themselves inside a quarter:
Quick Answer: The mix differs sharply by engine. Google’s AI features lean hardest on your own site and business profile. ChatGPT and Perplexity spread wider, pulling far more from forums, directories and third-party write-ups.
The grid below shows citation share by source type across the Malaysian prompt sets we track.
| Source type | Google AI features | ChatGPT | Perplexity |
|---|---|---|---|
| Business’s own website | 41% | 27% | 30% |
| Business profile and map listings | 23% | 11% | 9% |
| Directories and category listings | 14% | 21% | 24% |
| Forums and community threads | 9% | 25% | 22% |
| News, blogs and third-party write-ups | 13% | 16% | 15% |
Source: ZenWeb citation tracking, Malaysian client prompt sets, 2025–2026. Licence.
The read: if your buyers use Google, your own pages and profile do most of the work. If they use ChatGPT, half your visibility sits on properties you do not control. That is why the off-site fortnight is not optional.
Quick Answer: The final fortnight compares week twelve against your week-one baseline. Three numbers matter: how often you are named, how often you are cited, and whether any of it produced an enquiry.
Measurement is far more workable than a year ago. Google now publishes a generative AI performance report in Search Console, so AI performance is no longer purely inferred.
Three measurements, three different tools:
Then decide. Keep the stages that moved something, drop the rest, and if the work needs a specialist rather than a retainer, read when a consultant beats a full agency before renewing.
Quick Answer: A typical Malaysian SME site absorbs roughly 90 to 110 hours across the quarter. Content rewriting takes the largest share by far, while the access work protecting everything else takes the least.
The chart below shows median hours per workstream on a 20 to 40 page site.
| Workstream | Share of effort | Hours | Stage |
|---|---|---|---|
| Content rewriting | 44 | 31–50 | |
| Off-site evidence | 21 | 51–70 | |
| Entity and identity | 17 | 15–30 | |
| Measurement setup | 13 | 71–90 | |
| Access and crawlability | 9 | 1–14 |
Source: ZenWeb operational data, Malaysian SME engagements, 2024–2026. Licence.
Nine hours protects the other ninety-five. That ratio is the argument for sequencing an AI search optimisation plan, and the number to quote when someone wants to start with the rewrite.
Want this quarter run properly the first time?
We sequence the work, hold the baseline and report against it monthly. Compare our SEO service scope →
Quick Answer: Skip anything sold as an AI-only technique. Google states plainly that llms.txt files, forced chunking, AI-specific rewriting and special AI schema are not used by its generative features. Those hours belong elsewhere.
Three things get proposed constantly and belong in nobody’s first quarter:
Skip the tooling debate too. No third-party tool sees any engine’s internal systems. For the honest version, the 15 steps that get you cited by chatbots covers it without the folklore.
Quick Answer: Access, then identity, then content, then off-site evidence, then measurement. Five stages, ninety days, one decision at the end about what to keep paying for.
An AI search optimisation plan is not a secret set of techniques. It is ordinary search work done in an order that lets you tell what worked. Most Malaysian SMEs find the first two stages fix more than expected, and that the third is where the real hours go.
Weighing in-house against outside help? The scope question is answered in our SEO services overview. Either way, take the baseline before you change anything — you only get one week zero.
Access fixes register within two weeks. Being named more often in answers typically starts moving around week eight, and referral traffic later still. A flat first month is normal even when the work is correct, which is why the week-one baseline matters more than weekly checks.
No. Google’s guidance says optimising for generative AI features is still SEO, rooted in the same ranking systems. What changes is the measurement layer and the emphasis on entity clarity and quotable answers. Treat an AI search optimisation plan as a re-sequencing of your SEO budget, not a second line item.
You can, but do not expect visibility from it. Google states that it does not use llms.txt or similar files, and that publishing one neither helps nor harms rankings. Keep it if another service reads it; otherwise spend the time on access checks and page clarity.
Yes, particularly in narrow local categories where few businesses have clear, well-described pages. The harder part is that a large share of ChatGPT citations come from directories and forums, so a small business has to earn accurate mentions off its own domain too.
Run the access check. Confirm retrieval crawlers are not blocked at robots.txt or your CDN, that your main pages are indexed, and that your content exists without JavaScript. It is the cheapest stage of any AI search optimisation plan and it protects every hour you spend afterwards.
Ready to be the business AI names?
Book a free 30-minute strategy session — we’ll check your crawler access, entity signals and how often the engines name you today, then give you a concrete 90-day plan.
Complete the form and our team will contact you to discuss your goals. Let’s grow your business.

Online