Most GEO pitches arrive with new vocabulary and an old promise. The vocabulary is generative engine optimisation, answer engine optimisation, AI visibility. The promise is that a monthly fee will place your brand inside AI answers. Knowing how to choose a GEO agency now depends less on judging the promise and more on checking the scope behind it.
That check became far easier in 2026. Google published two documents aimed squarely at buyers: a guide to optimising for generative AI features on Search, and separate guidance on third-party SEO tools and services. Read together, they hand you a free vetting standard.
We should declare our position. IZI Digital Marketing sells this work, so treat the criteria below as testable claims rather than as neutral commentary. Apply them to us as strictly as to anyone else. The market context is real enough: ICT and e-commerce activity made up 23.5 per cent of Malaysia’s GDP in 2023, worth RM427.7 billion, according to the Department of Statistics Malaysia. Plenty of budget is moving. The question is what a GEO and AEO programme can honestly buy with it.
SEO, AIO, GEO, your site, & third-party support to optimize for LLMs
Source video: Google Search Central on YouTube
PART 1 · DIAGNOSE
What Can a GEO Agency Actually Control?
IN BRIEFA GEO agency controls whether your pages are retrievable, credible and worth citing. It does not control the model. On Google surfaces the same index decides both ordinary results and AI answers, which is why AI SEO and traditional SEO should be funded as one programme.
Google describes two mechanics behind its AI answers. Retrieval-augmented generation grounds the response in pages pulled from the ordinary Search index. Query fan-out means the model quietly issues related queries and gathers extra results before answering.
Both mechanics point the same way. Influence sits upstream of the answer, in whether your page is indexed, eligible for a snippet, and better than the alternatives on the same question.
- Retrieval is the gate. A page must meet the Search technical requirements and be snippet-eligible before any AI feature can use it.
- Inclusion is a setting. A site excluded from generative AI features in Search Console cannot appear in them, whatever the retainer promises.
- Fan-out rewards depth. Answers get assembled from queries nobody typed, so thin single-keyword pages get bypassed.
- The model itself is off-limits. No agency can edit a model’s weights, and none has a back door into ranking.
Not sure whether your site is even eligible yet?
An eligibility check takes under an hour and usually reorders the whole shortlist. See what a proper audit covers
PART 2 · DIAGNOSE
What Does Google Say About Buying GEO Services?
IN BRIEFGoogle says it does not evaluate third-party services, that no external tool sees its internal ranking data, and that buyers should check any advice against official guidance. That single paragraph turns answer engine optimisation from a belief question into a documentation question.
This is the part of the decision most buyers skip. Google’s third-party guidance names the exact category: services “promising improvements for AI experiences and search formats (also known as ‘AEO’ or ‘GEO’ tools)”. Then it says something worth quoting in any procurement meeting.
Good advice, it says, either qualifies its claims as opinion based on data or experience, or backs them up by citing official Google Search guidance. That is a usable test, and it costs nothing to run.
- No approval exists to claim. Google does not evaluate third-party services, so “Google-approved method” is not a thing that can be true.
- No tool has internal data. Third-party tools cannot see ranking systems and cannot guarantee performance; predictions are their own.
- First-party data is free. Google recommends Search Console as the source that comes from Search itself.
None of that makes GEO agencies unnecessary. Google’s own guide on whether you need an SEO still treats skilled help as sensible. It just moves the burden of proof from the agency’s confidence to the agency’s citations.
PART 3 · DESIGN
Which Questions Separate a GEO Specialist From a Rebrand?
IN BRIEFSix questions, asked in order, usually settle it inside one meeting. A specialist answers with surfaces, evidence and failure conditions. A rebranded package answers with vocabulary. The same discipline applies to any retainer, as our list of questions to ask before signing sets out.
How to vet a GEO agency in one meeting
Run these in sequence. Each answer narrows what the next question can hide.
- Name the surfaces. Ask which surfaces are in scope (AI Overviews, AI Mode, ChatGPT, Perplexity) and how each one is measured separately. “AI search” as a single line item means the scope has not been written.
- Show the baseline. Ask what is being recorded in week one. Without a baseline, month three has nothing to compare against and every result becomes a matter of opinion.
- Trace the guidance. Ask which official Google page supports each tactic. Then read the mythbusting section of Google’s guide together and see how much of the plan survives it.
- Separate the work from the words. Ask what physically changes on the site. If the answer is only files and markup, ask what a human reader gains from the month.
- Meet the actual team. Ask who does the work, not who presents it. The person in the pitch is often not the person on the account.
- Define failure. Ask what evidence at month six would mean the programme failed. An agency that cannot name a failure condition cannot be held to one.
BENCHMARK BRIEFING 1 OF 4
Which GEO Tactics Survive Google’s Own Guidance?
IN BRIEFSeveral headline GEO tactics are explicitly listed by Google as unnecessary for its generative AI features. The table maps seven common proposal line items against what Google has published, so a buyer can score a scope document line by line before signing anything.
| Proposal line item | Google’s published position | What to do with it |
|---|---|---|
| llms.txt file | Not used by Google Search; neither helps nor harms | Fine to keep, never a billable deliverable |
| Chunking content for AI | No requirement; no ideal page length | Remove from scope; reformat for readers instead |
| Rewriting content for AI systems | Not needed; systems understand synonyms and meaning | Redirect the hours into original reporting |
| Buying “AI mentions” across the web | Inauthentic mentions are not as helpful as they seem | Decline; it also risks spam systems |
| Special AI schema markup | No special schema required for generative AI features | Keep schema for rich results, not as a GEO claim |
| Indexing and snippet eligibility | A precondition for appearing in AI features | Insist it is the first month’s work |
| Unique, non-commodity content | Likely to influence presence more than anything else | Give it the largest share of the retainer |
Source: compiled from Google Search Central documentation, 2026. Licence.
PART 4 · DESIGN
Specialist GEO Agency, Full-Service Partner, or In-House?
IN BRIEFThree routes exist, and the deciding factor is usually who controls the website, not who knows the most about AI search. Where nobody can ship a site change quickly, a specialist stalls. We compare the trade-offs in DIY AI SEO versus hiring a specialist.
Most Malaysian SMEs treat this as a knowledge question. It is closer to a plumbing question: whichever route can get an approved change published within a week will out-perform the route with the better slide deck.
DECISION BOX · WHO SHOULD RUN YOUR GEO WORK
| Option | Best fit | Speed to first change | Main risk |
|---|---|---|---|
| Specialist GEO agency | Solid site, weak content depth | Fast, 2–4 weeks | Blocked by your dev queue |
| Full-service SEO partner | Site, content and search all weak | Moderate, 4–8 weeks | GEO becomes a small line item |
| In-house with advisory | Subject expertise already inside | Slow to start, fast after | Stalls when the owner gets busy |
Verdict: Choose a specialist only if someone on your side can approve and publish site changes weekly. If not, buy the full-service route, or fix the publishing bottleneck first and revisit the shortlist in a quarter.
Torn between a specialist and a full-service partner?
One session usually settles it, because the answer sits in your publishing workflow rather than in either proposal. See how the Blueprint sessions run
BENCHMARK BRIEFING 2 OF 4
Where Does GEO Evidence Actually Come From?
IN BRIEFOnly some GEO evidence is verifiable by the client. The grid below maps five common report lines against three sources, showing which numbers you can check yourself in an account you own. That is the test that matters when comparing AI visibility tracking tools.
| Report line | Search Console | Your analytics | Third-party tool |
|---|---|---|---|
| Impressions in generative AI features | Yes, first-party | No | Estimated only |
| Clicks from AI answers | Not in the AI view | Partly, by referrer | No |
| Which page was surfaced | Yes, by page | Only if clicked | Sampled |
| Citations in non-Google assistants | No | Referral sessions only | Yes, prompt-sampled |
| Enquiries and revenue | No | Yes, if tracked | No |
Source: compiled from Google Search Central and platform documentation, 2026.
PART 5 · DEPLOY
What Should the First 90 Days With a GEO Agency Look Like?
IN BRIEFWeeks one to four go on eligibility and baseline, weeks five to eight on the pages that answer real buying questions, weeks nine to twelve on measurement. If a plan front-loads publishing volume instead, the agency is buying time, not visibility.
The sequence matters more than the activity list, because each phase makes the next one measurable. A programme that starts by publishing has no way of proving what the publishing did.
- Weeks 1–4: eligibility and baseline. Confirm indexing, snippet eligibility and inclusion settings, then record where you currently appear so month three has a reference point.
- Weeks 5–8: depth where it pays. Rebuild the handful of pages that answer the questions your buyers actually ask, with the helpful, people-first standard as the bar.
- Weeks 9–12: measurement and review. Compare the AI-feature view in Search Console against enquiries, and decide what changes next quarter.
Note what is absent. There is no month of file creation, no bulk publishing sprint, and no promise of a position inside an answer. Volume for its own sake also runs into Google’s scaled content abuse policy, which is a poor way to spend a first quarter.
BENCHMARK BRIEFING 3 OF 4
How Much Weight Should Each Selection Criterion Carry?
IN BRIEFMost shortlists collapse because price is weighted first. This illustrative scoring model puts evidence and scope ahead of commercial terms, which is the ordering that survives contact with a month-by-month ROI review.
| Criterion | Relative weight | Points |
|---|---|---|
| Evidence and reporting | 30 | |
| Scope clarity by surface | 25 | |
| Technical foundations | 20 | |
| Content capability | 15 | |
| Commercial terms | 10 |
Illustrative model by IZI Digital Marketing, built on Google Search Central guidance, 2026. Licence.
PART 6 · DRIVE
How Do You Read the Month-Three Report?
IN BRIEFRead it backwards. Start at enquiries, then look for the pages behind them, and only then read the AI-visibility scores. Reports that open with proprietary scores and end with business outcomes are arranged to flatter, a pattern our note on agency red flags covers in more detail.
Three months is long enough to judge the process and too short to judge the outcome. So grade the process honestly and hold the outcome open.
- Baseline present and unchanged. If the starting numbers have been quietly restated, treat every comparison in the report as unreliable.
- Named surfaces, separate lines. Google features and other assistants move for different reasons and should never be reported as one blended score.
- Work shipped, not work recommended. Count what actually went live, and whose approval stalled the rest.
- A written call. Continue, adjust or stop — with the reason. Silence at month three usually becomes an awkward month nine.
Got a GEO proposal on the table right now?
We will read it against Google’s published guidance line by line and tell you which parts are billable work. Check the contract clauses first
BENCHMARK BRIEFING 4 OF 4
What Has Google Published on GEO, and When?
IN BRIEFFour dated releases in 2026 built the standard a buyer can now hold an agency to. The timeline below tracks what each one published and what it lets you verify, which is why the AI SEO question has become answerable rather than speculative.
| Release | Jan 2026 | May 2026 | Jun 2026 | Jul 2026 |
|---|---|---|---|---|
| What Google published | Search Off the Record on hiring and third-party tools | Generative AI optimisation guide, with mythbusting | Generative AI performance reports in Search Console | Guide refreshed, agentic browsing section added |
| What a buyer can verify | That no tool holds internal ranking data | Which proposal items Google calls unnecessary | Impressions in AI features, by page and country | Whether the scope reflects current guidance |
Source: compiled from Search Console Help and Google Search Central, Jan–Jul 2026.
THE VERDICT
So How Should You Choose a GEO Agency?
Choose the agency whose proposal you can check. That is the whole standard, and 2026 is the first year it is genuinely available to a buyer without technical training, because the guidance, the mythbusting list and the first-party reporting all exist in public.
Practically, that means three things. Score evidence and scope ahead of price. Insist the first quarter starts with eligibility and a baseline. Keep the success criteria tied to numbers in accounts you own. If a shortlist is still tied after that, our comparison of AI SEO agencies for GEO work and the wider Malaysian SEO agency shortlist set out how we would break the tie. Our AI SEO, GEO and AEO service page shows what we put in scope ourselves.
FAQ
Frequently Asked Questions
1. Is a GEO agency different from an SEO agency?
In practice, rarely. From Google Search’s perspective, optimising for generative AI features is optimising for search, so the underlying work overlaps almost completely. The difference that matters is measurement: a GEO-capable team reports AI-feature impressions and assistant citations separately instead of blending everything into one traffic line.
2. How much should I pay for GEO work in Malaysia?
Pay for the work, not the label. The right figure depends on how much of your site needs technical repair and how many pages genuinely need rewriting. A site already indexed and well structured needs a far smaller programme than one that is not. Ask any quote to be split by phase so you can see what you are funding.
3. Can any agency guarantee my brand appears in AI Overviews?
No, and a guarantee is a reason to walk away. Google states that indexing and serving are never guaranteed, and that no third-party tool has access to its ranking systems. Agencies can raise the probability by making pages retrievable and genuinely better; nobody can promise placement inside a generated answer.
4. Do I need llms.txt or special AI schema to be cited?
Not for Google Search. Google says llms.txt and similar files are ignored by Search, and that no special schema is required for its generative AI features. Keep structured data for rich-result eligibility, but treat either item as a billable GEO deliverable and you are paying for something Google has published as unnecessary.
5. How long before a GEO programme shows anything?
Expect signal at three months and judgement at six. The pace depends mostly on how fast approved changes actually get published: a weekly publishing rhythm compounds, a fortnightly approval queue does not. Set the review date when you sign, not when the results feel disappointing.
Want a second opinion before you sign a GEO retainer?
Book a free Blueprint consultation — we will read the proposal against Google’s published guidance, tell you which lines are real work, and hand you the baseline and review dates to hold whoever you hire to.