GEO Case Study: How We Got a Singapore Brand Cited in ChatGPT

First published: 20 July 2026 · Last updated: 20 July 2026

1

Prompt audit

Map the 50 to 100 prompts your real buyers actually ask ChatGPT, Perplexity and Gemini. Stop optimising for keywords. Optimise for prompts.

2

Page restructure

Answer blocks (40-60 words) on every priority page, comprehensive topical coverage, schema completeness, structured comparison and FAQ blocks.

3

Third-party SG citations

Earn mentions on SG-specific publications AI engines weight heavily (Vulcan Post, MoneySmart, SETHLUI, niche industry blogs). Distribution beats domain authority.

4

Brand entity lockdown

Wikidata, Crunchbase, LinkedIn, Google Business, your site footer all carrying identical brand name, founding date, category, address. Inconsistency kills citations.

This is one of the most-asked questions we get from SG marketing leads in 2026: "Can you actually get our brand cited in ChatGPT, or is GEO just SEO with a new label?" The honest answer is yes, you can, and no, it is not the same as SEO. The methodology overlaps significantly with technical SEO foundations but the targeting, structure and measurement are different. This piece walks through one anonymised SG case study end-to-end so you can judge for yourself. The brand is a mid-tier B2C service business operating only in Singapore. Roughly SGD 4M annual revenue, 12 staff, 4 years old. Strong brand within their niche but unknown outside it. Decent SEO baseline (DR 28, ranking page 1 for branded queries and a handful of unbranded long-tails). Zero ChatGPT or Perplexity visibility at the start of the engagement. We have anonymised the brand for client confidentiality but every number, methodology step and timeline below is unedited. For deeper technical context on this work, our sister site BestSEO published the practitioner-focused version of this case study at the BestSEO ChatGPT citation deep-dive; that piece covers the schema, technical SEO and content engineering in more depth. For the foundational "what is GEO" context, see our explainer on what is generative engine optimization. For the broader AI-engine landscape, our Perplexity optimisation guide covers the parallel work for the second-largest AI search engine. For the citation-worthy content patterns we used heavily here, see our anatomy of citation-worthy content.

The Starting Position

Before we started, we ran a baseline audit across the major AI engines. Our protocol:

• 50 SG-relevant prompts identified through customer interviews and search-console mining
• Each prompt tested across ChatGPT (GPT-5), Perplexity (Pro), Gemini (2.5 Pro), Google AI Overviews
• Brand mention and citation tracking via Profound (with manual verification on 20% sample)
• Run once per week to control for output variance

**Baseline results, week 0:**

• ChatGPT: 0 citations across 50 prompts. Brand mentioned in passing 3 times, never as a recommended option.
• Perplexity: 2 citations across 50 prompts. Both incidental (brand appeared in a list, no link emphasis).
• Gemini: 1 citation across 50 prompts. The brand's own homepage as a single source.
• Google AI Overviews: 4 appearances across 50 prompts. Mostly branded queries.

This is roughly the AI visibility profile of every SG SME we audit who has not done explicit GEO/AEO work. Strong on branded queries, invisible on unbranded category queries. The brand's ranking on Google was reasonable (page 1 for ~40 of the 50 prompt-equivalent keywords) but the AI engines were citing competitors with stronger entity authority and better-structured content.

The 50-Prompt Map

The single most important step we took was building the prompt map. Most GEO work fails at the first step because teams optimise for keywords instead of prompts. Keywords are the search query a user types into Google. Prompts are the longer, conversational, multi-criteria questions a user asks ChatGPT. Compare:

• Keyword: "wedding planner singapore"
• Prompt: "I am getting married in October, budget around 25k for the planner role only, want someone who can handle a 200-pax indoor venue and is good with non-Chinese rituals. Who should I look at in Singapore?"

The keyword has been competed for by every SG wedding planner for a decade. The prompt is fresh territory. The AI engine extracts criteria (October, 25k budget, planner-only scope, 200-pax indoor, non-Chinese rituals) and surfaces 2 to 4 named recommendations. The brand that wins the keyword is not necessarily the brand that wins the prompt. Our prompt-mapping process for this client:

1. Customer interviews (8 interviews, 30 minutes each, recent customers and lost prospects). Asked them to recreate the actual question they would ask ChatGPT about the brand's category. Captured exact phrasing.
2. Search Console mining for long-tail informational queries the site already ranked for, then expanded each into a "would a user ask this as a chat prompt" version.
3. AI engine probing. Took the seed prompts and asked ChatGPT itself "what are 10 variations of this question that real users might ask". Then trimmed for SG specificity.
4. Competitor citation reverse-engineering. Ran our top 5 competitors' brand names through ChatGPT with "what should I know about [brand]". Captured the contexts the AI engine surfaced them in. Reverse-engineered the prompts that would surface them.

End output: a Google Sheet with 50 prompts, each tagged by buyer-journey stage (top, middle, bottom of funnel), criteria mentioned (budget, location, use case, demographic), and current AI engine response (which brands cited, which sources linked). This sheet became the optimisation target for everything else. We were no longer trying to "rank for keywords". We were trying to "be the recommended answer to these 50 prompts".

Element	Keyword (SEO target)	Prompt (GEO target)
Length	2-5 words	20-80 words
Specificity	Low	High (multiple criteria)
Audience qualifier	Implicit	Explicit (budget, location, demographic, use case)
Output format	10 blue links	2-4 named recommendations + 3-7 sources
Optimisation target	Rank position	Citation in answer
Competition density	Saturated	Mostly empty (in 2026)
SG opportunity	Hard to win	Underexploited

Move 1: Page Restructure (Weeks 1-4)

With the prompt map in hand, we restructured the brand's top 30 pages (homepage, top 5 service pages, top 8 location pages, top 16 informational blog posts that ranked for prompt-adjacent keywords).

Per page, the restructure included:

• Answer block of 40 to 60 words at the top, directly under H1, written to literally answer the dominant prompt the page targets. Plain language, self-contained, includes the primary "answer terms" the AI engine would lift.
• Comprehensive topical coverage. Every page expanded to cover the full set of buyer-relevant criteria the prompt map identified. If users asked about price, the page named price ranges. If users asked about timeline, the page stated typical timelines. Vague pages get skipped by AI engines.
• Schema completeness. Service schema on service pages, FAQPage schema on every FAQ block, Article schema on blog posts, Organization schema with sameAs links to LinkedIn, Wikidata, Crunchbase. Validated through Schema.org validator and Google Rich Results Test.
• Structured comparison content. On service pages, added "How we compare to alternatives" sections that named the three or four real SG competitors and graded honestly across price, scope, ideal-client fit, and limitation. Included our brand in the comparison without pretending to win every dimension.
• FAQ blocks built from prompt map. Every page got a FAQ section of 5 to 8 questions drawn from the 50-prompt sheet, with FAQPage schema.

Output by end of week 4:

• 30 pages restructured
• 30 answer blocks live, each 40 to 60 words, validated
• ~180 FAQ entries added, all schema-tagged
• 30 service/article schema instances validated
• Site-wide Organization schema with consistent entity data

Re-ran the citation audit at week 4. ChatGPT citations: 6 of 50 prompts (up from 0). Perplexity: 8 of 50 (up from 2). Gemini: 4 of 50 (up from 1). Early signal that the work was moving the needle.

Move 2: Third-Party SG Citation Building (Weeks 3-10)

Page restructure alone gets you maybe 10 to 15% of the way to strong AI citation visibility. The bigger lever is third-party citations, ideally on SG-specific publications that AI engines weight heavily for SG queries.

The publications that consistently move the needle for SG B2C brands in our experience:

• Vulcan Post (SG startup and lifestyle)
• MoneySmart blog (SG financial services)
• SETHLUI (SG food and lifestyle)
• MotherShip (SG general lifestyle)
• The Smart Local (SG lifestyle and travel)
• HoneyKids Asia (SG family and parenting)
• Niche industry-specific SG blogs (Hardwarezone for tech, sgCarMart blog for auto, etc)

For B2B brands the list shifts to The Business Times, Tech in Asia (SG edition), SBR (Singapore Business Review), e27, Tech Wire Asia, niche LinkedIn newsletters.

Our outreach approach for this client:

1. Identified 12 target publications matching the brand's category and audience.
2. Pitched 3 angles per publication based on what each publication had already covered in the past 6 months. We pitched genuine value (data, expert commentary, founder story) rather than promotional features.
3. Offered exclusive data or commentary drawn from the brand's own customer base (anonymised). Publications love proprietary data.
4. Followed up twice per pitch. Standard PR cadence.

Outcomes over weeks 3-10:

• 7 of 12 publications said yes
• 9 published mentions secured (some pubs ran two pieces)
• 4 of these 9 included direct backlinks; 5 were brand mentions without links (still valuable for AI citation; AI engines pick up name mentions, not just hyperlinks)
• Total spend on PR outreach (writer fees, light retainer for senior PR contractor): SGD 6,500 over 8 weeks

Re-ran the citation audit at week 10. ChatGPT: 19 of 50 prompts. Perplexity: 22 of 50. Gemini: 11 of 50. The third-party citations moved the needle significantly more than the page restructure alone.

This matches what AEO practitioners now widely report: third-party content distribution can lift AI citation rates by 200 to 300% versus owned content alone. Distribution beats domain authority.

Move 3: Brand Entity Lockdown (Weeks 4-8, in parallel)

The third move is the one that most SG brands skip and the one that quietly amplifies everything else. AI engines do not just look at content; they look at the entity behind the content. If your brand is described as "ABC Pte Ltd" on your site, "ABC Singapore" on LinkedIn, "ABC Holdings" on Crunchbase, and not present at all on Wikidata, the AI engine sees four ambiguous entities and downweights all of them.

Our entity-lockdown checklist for this client:

• Site footer and About page: standardised brand name, founding date, registered address, UEN, category, social links.
• Schema Organization markup: consistent name, foundingDate, address, telephone, sameAs links to all owned profiles.
• Google Business Profile: updated category, services, photos, posts, Q&A. Maintained 4.7+ rating with active review responses.
• LinkedIn Company Page: matched brand name, founding date, category, About copy. Updated weekly with relevant posts to maintain freshness.
• Crunchbase: claimed and updated profile with consistent data, founding date, headcount, category, descriptions.
• Wikidata entry: created (this brand had no Wikidata presence). Added P31 (instance of: business), P17 (country: Singapore), P571 (inception date), P159 (headquarters location), P452 (industry), P856 (official website). Took 3 weeks to be approved and indexed.
• Wikipedia: evaluated; brand did not yet meet notability threshold. Skipped (forcing a Wikipedia article for a non-notable brand backfires).
• Industry-specific directories: updated 4 SG industry-specific directory profiles for the brand's category.

By week 8 the brand had a tight, consistent entity profile across all the sources AI engines crawl. The Wikidata entry alone, indexed by week 7, lifted brand recognition in AI engines noticeably. AI engines treat Wikidata as a high-trust entity source.

Source	AI weight	Effort to fix	Notes
Wikipedia	Highest	Hard (notability bar)	Force is counterproductive
Wikidata	Very high	Medium	Most underused; create even without Wikipedia
Google Business Profile	High	Easy	Active management matters
LinkedIn Company Page	High	Easy	Posting cadence affects freshness signal
Crunchbase	High	Medium	Claim and complete
Schema Organization on site	High	Easy	sameAs links to all profiles
Industry-specific directories	Medium	Easy-Medium	Vary by category
Press citations (SG)	High	Hard	Move 2 territory

Move 4: Comprehensive Content Plays (Weeks 5-12)

In parallel with citation outreach we built four flagship content pieces designed specifically to be cited as definitive sources in AI engines:

1. A "best [category] in Singapore" buyer guide that named 8 SG brands (including the client) and graded honestly across criteria. This is the format AI engines preferentially cite for "best X in Singapore" prompts.
2. An original SG-specific data study based on the brand's own anonymised customer data. ~1,200 word piece with 4 charts, distinctive proprietary stats. The kind of asset PR can pitch and AI engines treat as primary source.
3. A comprehensive "how to choose a [category provider] in Singapore" guide covering criteria, red flags, typical pricing, what to ask in a discovery call. Educational, no overt sales angle.
4. A SG-specific FAQ hub consolidating the top 30 questions buyers ask in the category, structured for quick-scan and machine-extractable.

These four pieces took roughly 60 hours of senior content production (combination of in-house and agency). They became the assets cited most often as the brand's AI citation profile lifted through weeks 8 to 12.

By week 12 the citation audit showed:

• ChatGPT: 28 of 50 prompts (56%)
• Perplexity: 31 of 50 prompts (62%)
• Gemini: 16 of 50 prompts (32%)
• Google AI Overviews: 14 of 50 prompts (28%)

This was a strong outcome. The brand was now being named, on most queries within their category, alongside (and sometimes above) competitors with significantly higher domain authority and budget.

Cost Breakdown

The total project cost over the 90-day engagement:

• Content production (page restructure, 4 flagship pieces, FAQ builds): SGD 9,500
• Technical work (schema implementation, Wikidata entry, validation): SGD 3,000
• Third-party citation outreach (PR contractor, writer fees): SGD 6,500
• Citation tracking (Profound subscription, manual auditing): SGD 1,500
• Project management and reporting: SGD 1,500

Total: SGD 22,000 over 90 days.

For an SG SME of this scale, this cost was roughly equivalent to one quarter of their existing paid social spend. The trade-off they made was to redirect 30% of paid budget into this GEO project for one quarter and assess the result.

Revenue Outcomes

We tracked AI-referred revenue in two ways:

1. GA4 referrer attribution. Sessions arriving from chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, etc. Imperfect because some AI engines do not pass referrer reliably, but a useful directional signal.
2. Customer survey at point of conversion. "How did you first hear about us?" with AI engines as named options.

Outcomes in the 90 days following project completion (Days 91 to 180):

• 412 GA4 sessions identified as AI-engine referred
• 47 customer enquiries (11% conversion rate, vs site average of 2.8%)
• 19 closed deals
• SGD 78,000 net new attributable revenue
• 3.5x ROI on the SGD 22,000 project cost in the first 90 days alone, with compounding tail expected as AI citations continue to drive enquiries

The conversion rate differential (11% vs 2.8% site average) is the pattern we see consistently in AI-referred traffic. Buyers who arrive having read the brand's name in an AI engine recommendation arrive significantly more pre-sold than buyers from cold organic or paid traffic.

What We Would Do Differently

In hindsight, three things we would change:

Start the entity lockdown in week 1, not week 4. Wikidata indexing took 3 weeks. If we had started immediately, the entity boost would have arrived earlier in the citation audit cycle.

Build the flagship content pieces first. We led with page restructure but the four flagship pieces drove more citations per asset. In future projects we sequence: prompt map (week 1), flagship pieces (weeks 1-4 in parallel with restructure), entity lockdown (week 1), citation outreach (week 3 onward).

Track Perplexity and Gemini outcomes with equal rigor as ChatGPT. The team focused most of the analytical attention on ChatGPT (largest engine). Perplexity actually delivered higher citation rate and is increasingly important for B2B and professional services queries.

What This Means for Other SG Brands

The methodology in this case study is not unique to this client. We have run variations of the SG Citation Stack across roughly 20 SG brands now. Outcomes vary by category, starting position and execution discipline, but the directional pattern is consistent: SG brands who do this work see meaningful AI citation visibility within 60 to 90 days, and AI-referred traffic converts at multiples of cold traffic.

What makes a brand a good fit for GEO investment in 2026:

• Some existing organic search baseline (DR 20+, ranking page 1 for branded queries at minimum)
• Real expertise in the category (you have something legitimately worth citing)
• Willingness to invest 60 to 90 days before judging outcomes
• Bandwidth to produce or commission flagship content
• Buyer journey that includes some research/comparison phase (not pure impulse purchase)

What does not fit:

• Brands with zero organic baseline (foundation work needed first)
• Categories where AI engines are not yet well-trained (very niche B2B verticals)
• Pure impulse-purchase brands where buyers do not research
• Brands unwilling to be honest about competitive comparison

For the broader category-level guidance on whether GEO is right for your business, see our GEO and AEO services page. For a foundational comparison with traditional SEO, see AEO vs SEO for Singapore businesses.

Frequently Asked Questions

How long does it take to get cited by ChatGPT?

For a brand with a reasonable SEO baseline (DR 20+, real expertise, decent existing content), realistic timeline: first citations on long-tail prompts within 30 days, meaningful citation rate (20%+ of tracked prompts) by Day 60, strong rate (50%+) by Day 90, compounding through Day 180. Brands starting from zero (no SEO baseline, thin content) need foundation work first and should expect a 6-month timeline to meaningful AI visibility. There is no "ChatGPT submission form"; citations are earned through the Citation Stack work above.

Is this just SEO with a different name?

No. The technical foundation overlaps significantly (schema, fast pages, structured content, internal linking). The strategic differences are: optimisation target is prompt-not-keyword, page structure prioritises answer blocks and comprehensive coverage over keyword density, third-party citations matter more than backlinks alone (mentions count too), brand entity consistency matters more, and measurement is citation rate and AI-referred conversion rather than rank position. Treat GEO as the next layer of SEO, not a separate discipline, but recognise the new layer requires new work.

Can I just buy ChatGPT advertising to get cited?

No. OpenAI does not currently sell paid placements within ChatGPT answers. Citations are earned through content authority, brand entity strength, and third-party mention density. Some experiments around sponsored placements have been floated but as of mid-2026 there is no "buy your way in" option for ChatGPT citations. This is one reason the window for organic-citation work is valuable: it will tighten when paid placements eventually arrive.

What does it cost to get cited by ChatGPT for an SG brand?

In the case study above, total cost was SGD 22,000 over 90 days for a comprehensive Citation Stack project on an SG mid-tier brand. Smaller projects (focused on fewer prompts and pages) can run SGD 8,000 to SGD 15,000 over 60 days and still produce meaningful results. Larger enterprise projects can run SGD 50,000+ over 6 months across multiple verticals and languages. The honest answer is "it scales with your prompt map size and existing baseline". For most SG SMEs, SGD 15K to 30K over 90 days is the sweet spot.

Which AI engine should I prioritise: ChatGPT, Perplexity, or Gemini?

Depends on your audience. For B2C and consumer queries, ChatGPT has the largest user base in SG and should be the primary target. For B2B, technical, and professional services queries, Perplexity over-indexes (its users are more research-heavy) and often delivers higher conversion. For Google ecosystem users, Gemini and Google AI Overviews matter increasingly. Most well-built Citation Stack work lifts visibility across all four engines simultaneously because the underlying signals (content, schema, entity, citations) are shared. Track all four; weight reporting to the engine your buyers actually use.

Can I do this without an agency?

Yes if you have a senior in-house marketer with 20+ hours per week to dedicate for 90 days, plus access to an experienced PR contractor for the citation outreach. The methodology is documented (see this article and the BestSEO technical deep-dive); the work is mechanical once you have the prompt map. Agencies move faster because they have done it 20 times and have established media relationships, but in-house is fully viable for SG SMEs with the bandwidth.

Related reading

• What is generative engine optimization (GEO) in Singapore
• How to optimise for Perplexity in Singapore
• The anatomy of citation-worthy content