First published: 20 May 2026 · Last updated: 20 May 2026
On-domain extractability
Quick Answer Boxes, FAQPage schema, structured tables, exact-match query inclusion in titles and H1s. The content shape ChatGPT lifts verbatim.
Off-domain brand mentions
Trade press features, named expert quotes in industry publications, government registry listings, podcast guesting. The trust signals ChatGPT's re-ranker weights heavily.
Bing index visibility
Bing Webmaster Tools verification, IndexNow, Bing-friendly schema, exact-match keyword discipline. ChatGPT Search retrieves through Bing's index.
The Brand and the Baseline
The brand: a Singapore-headquartered B2B SaaS company in the supply-chain visibility category. Approximately 50 employees, mid-7-figure ARR, primary market is Southeast Asia MNC procurement and operations teams. We have anonymised the brand and the specific category for client confidentiality, but the framework and the numbers are documented from the actual engagement. The starting position in October 2025:- ChatGPT Search citations: 0 across 50 tracked vendor-evaluation prompts.
- Perplexity citations: 2 (both as sub-citations within larger lists).
- Google AI Overviews citations: 4 across 50 prompts.
- Bing organic ranking: position 30+ on most target queries.
- Domain authority signals: moderate. Industry-recognised brand, mid-tier backlink profile, no Wikipedia entry, sparse trade press coverage.
Pillar 1: On-Domain Content Reshaping
The first 30 days focused on reshaping the existing top 12 priority pages for verbatim extraction. We did not create new pages. The hypothesis: ChatGPT Search's synthesis layer disproportionately lifts content shapes that match its output template (definition-first openers, question-answer pairs, structured comparisons). Reshaping existing pages with this template should raise extractability without adding new content debt. The reshape template applied per page:Pillar 2: Off-Domain Brand Mention Programme
Days 15 to 75 ran in parallel with the on-domain reshape and overlapped into the next phase. The hypothesis: ChatGPT Search's re-ranker explicitly weights brand mentions in trusted publications heavily, often more heavily than on-domain content quality. A brand cited in 4 to 6 trade press articles and 2 podcasts will out-cite a brand with no off-domain mentions, controlling for content quality. The off-domain interventions over 60 days: 1. Trade press placements. We pitched the brand's CEO and head of product to four ASEAN logistics and supply chain trade publications. Two confirmed and published feature pieces (one trend commentary, one Q&A interview) within the 60-day window. The pieces named the brand by name in the headline or first paragraph and linked to the brand site. 2. Podcast guesting. The CEO recorded three podcast episodes over 45 days, two with logistics industry podcasts and one with a Singapore B2B SaaS founders podcast. All three published within the 90-day window. Show notes included the brand name and a backlink. Podcast transcripts were submitted via the host's site for indexing. 3. Industry research contribution. We packaged a small piece of original anonymised research from the brand's product (port congestion patterns across 2 specific Southeast Asian ports over Q4 2025) and pitched it to three industry analyst publications. One published, citing the brand as the data source. 4. Wikipedia and Wikidata eligibility. The brand did not yet meet Wikipedia notability standards (independent secondary source coverage), so we focused on Wikidata first. Created a Wikidata entity for the company with structured properties (founded date, headquarters location, industry, founder names, official website). Wikidata is a primary citation source for ChatGPT Search entity disambiguation. 5. Government registry visibility. ACRA registration was already in order. We added the brand to two relevant SG industry registries (Singapore Logistics Association directory, Enterprise Singapore digital exporter registry). Both are crawled and weighted by ChatGPT's trust scoring for SG entity verification. 6. Named expert quotes pitched to existing articles. We tracked HARO and Qwoted-equivalent SG sources and pitched the CEO as an expert source for 8 in-progress journalist queries over 30 days. Four were accepted and published in trade publications, with the brand named. The total off-domain output by day 75: 2 trade press features, 3 podcast episodes, 1 industry research citation, 1 Wikidata entity, 2 industry registry listings, 4 named expert quote placements. Approximately 25 hours of focused PR work distributed across the CEO, head of product, and an external PR consultant.Pillar 3: Bing Visibility Programme
In parallel from day 1, the Bing programme ran through day 90. The hypothesis: ChatGPT Search retrieves through Bing's index before applying its GPT-4o re-ranker. A page that does not rank in Bing's top 20 for a target query will not enter ChatGPT's re-ranking consideration set. The Bing-specific interventions: 1. Bing Webmaster Tools full setup with sitemap submission, GSC import, and IndexNow activation. Two-hour setup once. IndexNow integrated into the Webflow CMS via a custom webhook, firing on every publish and update. 2. Schema validation through BWT URL Inspection on all 12 priority pages. Two pages had FAQPage schema that validated in Rich Results Test but failed parsing in Bing. Fixed (Bing was stricter on the `mainEntity` array structure). 3. Exact-match keyword discipline in title tags and H1s on the 12 priority pages. Bing's literal matching meant that semantic-equivalent phrases (synonyms, paraphrases) were costing 5 to 10 ranking positions. Tightened title tags and H1s to include the exact target query. 4. LinkedIn distribution workflow for every reshape and every off-domain placement. The CEO posted on LinkedIn within 24 hours of each milestone, tagging the brand company page and the publishing source where applicable. LinkedIn engagement is a Bing ranking signal (and Microsoft owns LinkedIn). 5. Bing-specific backlink targets. Used Bing Webmaster Tools Backlinks report to identify Bing-indexed backlink opportunities not visible in GSC. Pitched 6 outreach targets, 2 converted. Both linking domains had stronger Bing-side authority than Google-side. By day 60, Bing organic rankings on the 12 priority pages had moved from average position 30+ to average position 8. By day 90, average position 5, with 4 of the 12 pages in position 1 to 3. For the full Bing AEO methodology, our AI crawlers piece covers the bot-management side that complements this work.The Weekly Measurement Discipline
Every Friday, we ran the same 50 prompts in ChatGPT, Perplexity, Gemini, and Google AI Overviews. We recorded for each:- Presence/absence of the brand in the response.
- The cited URL if cited.
- The position of the citation in the response (top citation, middle, bottom of list).
- The exact language used (verbatim brand mention, paraphrased, generic category mention).
The shape: zero movement through week 3 despite the on-domain reshape going live. First citation in week 4 attributed to a single prompt where the on-domain reshape alone was enough to move the needle (a mid-tail comparison query). The inflection point at week 8 to week 10 corresponds to when the trade press placements and podcast episodes published and were indexed. The compounding through week 12 reflects the accumulation of off-domain trust signals being weighted by ChatGPT's re-ranker.
What Moved the Needle: Per-Tactic Attribution
By day 90, we had enough citation events (18) to attribute roughly which tactic moved which citation. The attribution is approximate because tactics compounded, but the pattern is informative.
| Tactic Cluster | Citations Attributed | Citation Type |
|---|---|---|
| On-domain reshape only (Pillar 1) | 4 | Mid-tail comparison and feature queries |
| Trade press features (Pillar 2) | 5 | High-intent vendor evaluation queries; trade press URLs co-cited |
| Podcast episodes (Pillar 2) | 2 | Founder-name and brand-name queries; podcast URLs co-cited |
| Industry research citation (Pillar 2) | 2 | Data-driven queries about port congestion specifically |
| Wikidata entity (Pillar 2) | 1 | Brand entity disambiguation in generic queries |
| Bing ranking improvement (Pillar 3) | 4 | Long-tail queries where retrieval-only was the bottleneck |
Two patterns are notable:
Off-domain trust signals dominated for high-intent queries. Trade press features and podcast episodes accounted for 7 of the 18 citations and disproportionately on the highest-value vendor evaluation queries. The brand-mention research finding from secondary sources (52 percent of cited pages featured original data or branded-owned insight) tracks our case study directly.
On-domain reshape was necessary but rarely sufficient. Only 4 citations were attributable to on-domain work alone. However, every off-domain win required the on-domain page to be in shape for ChatGPT to extract from when the user clicked through. Without the reshape, the trade press traffic would have arrived but the citation eligibility would not have compounded.
What We Would Do Differently
Three honest revisions for the next case study:
1. Start the off-domain programme earlier. We sequenced on-domain first, off-domain starting day 15. In hindsight, both should have started day 1. Trade press has a long lead time (pitch to publish averages 30 to 60 days) and starting earlier would have moved the inflection point from week 8 forward to week 5 or 6.
2. Run the prompt set across more engines, not more prompts. We ran 50 prompts in 4 engines weekly. We should have run the same 50 prompts in 6 engines (adding Claude and DuckDuckGo Browse) to capture the full multi-engine picture. Adding more prompts would have diluted measurement precision per prompt; adding more engines would have surfaced where the brand was over-indexed or under-indexed by engine.
3. Build the measurement dashboard from week 1. We tracked in a Google Sheet for 12 weeks. By week 6 the sheet was unwieldy. A simple custom dashboard (or even a paid tool like Otterly.AI for the tracking layer) would have saved roughly 10 hours of manual tabulation across the 90 days.
For the broader EEAT context that underpins all three pillars, our 2026 EEAT guide covers how the September 2025 Quality Rater Guidelines update interacts with AI engine trust scoring. The same trust signals that move QRG ratings move ChatGPT re-ranker decisions.
The Reproducibility Test
The framework is reproducible for any Singapore B2B brand on a similar resource budget (60 hours, 90 days, no paid AI tracking tool). The reproducibility caveats:
- The brand needs a real product story. We could pitch trade press features because the brand had genuine novelty in their port congestion data and a founder with strong industry credentials. A commodity SaaS with no differentiated angle would struggle on the off-domain pillar.
- The category needs Bing visibility headroom. B2B categories where MNCs are the buyer have lower Bing competitive density, leaving room to move from position 30 to position 5 in 60 days. B2C categories with global competition (consumer fitness apps, generic ecommerce) face much higher Bing competitive density and the same Bing programme produces less ranking lift.
- The CEO needs to be available for off-domain work. Trade press features, podcast guesting, and named expert quotes all require the founder or a senior named spokesperson. Brands without this internal capacity need to either build it or hire fractional spokesperson capacity, both of which extend the timeline.
For brands that meet the reproducibility criteria, the framework should produce broadly comparable results in 90 days. For brands that do not, the framework still applies but the timeline stretches to 120 to 180 days and the off-domain pillar requires more outreach volume to convert.
For the underlying GEO tactic library, our GEO playbook documents 9 individual tactics that map to the three pillars in this case study.
Frequently Asked Questions
Why do brands need three concurrent pillars to get cited by ChatGPT Search?
ChatGPT Search applies retrieval, re-ranking, and synthesis in sequence. Bing visibility (Pillar 3) controls retrieval: a page that is not in Bing's top 20 for a target query never enters the re-ranking consideration set. Off-domain brand mentions (Pillar 2) drive re-ranking: ChatGPT's re-ranker weights trust signals from publications it considers authoritative more heavily than on-domain content quality alone. On-domain content shape (Pillar 1) drives synthesis: pages with definition-first openers, FAQPage schema, and structured tables are extracted verbatim into responses more often than equivalent unstructured content. Drop any pillar and citation count plateaus because you have created a bottleneck in one stage of the pipeline.
How long does it typically take to see ChatGPT Search citations after starting a programme?
In the documented case study, first citation appeared in week 4, the inflection point was week 8 to 10, and material share-of-voice movement (above 20 percent) was visible by day 90. The lag pattern matters: ChatGPT Search re-indexes content with a 7 to 21 day delay, off-domain trust signals take 30 to 60 days from pitch to publication, and the re-ranker requires multiple compounding signals to move citation propensity. Budget 90 days minimum for material lift. Brands without a real product story or with high-competition B2C categories typically need 120 to 180 days.
How do you measure ChatGPT Search citations without a paid tool?
Manual prompt tracking. Identify 30 to 50 prompts a real buyer of your product would type. Run each prompt weekly in ChatGPT, Perplexity, Gemini, and Google AI Overviews. Record presence/absence, cited URL, position of citation in the response, and the exact language used. Two hours per week per 50 prompts. The data quality is high because you are reading every response in full. Paid tools (Otterly.AI, Profound, BrightEdge AI) automate the tracking and scale to 200+ prompts per engine, but for a 50-prompt tracking discipline, manual is sufficient and forces qualitative observation that automated tools do not surface.
Which pillar matters most if a brand can only invest in one?
Pillar 2 (off-domain trust signals) for high-intent queries. In the case study, trade press features and podcast episodes accounted for 7 of the 18 citations and disproportionately on the highest-value vendor evaluation queries. However, no single pillar produces material citation lift in isolation. A brand with extensive trade press coverage but a poorly structured website will still under-cite because ChatGPT cannot extract the answer cleanly. A brand with perfect on-domain content but no off-domain trust signals will be skipped by the re-ranker. The honest answer is that all three are necessary for material lift; the question is one of sequencing rather than substitution.
Does this framework work for B2C brands or only B2B?
The framework applies to both, but the timeline and pillar weighting differ. B2B brands targeting MNC procurement (the case study) benefit from lower Bing competitive density and from Microsoft Copilot enterprise penetration, both of which accelerate Pillar 3 results. B2C brands face higher Bing competitive density and rely more heavily on Pillar 1 (on-domain content shape) and Pillar 2 (consumer review platforms, lifestyle press, influencer mentions). For B2C, Wikipedia notability is more achievable and more impactful, and earned media targets shift from trade press to consumer publications. Budget 120 to 180 days for B2C versus 90 days for B2B with the same intervention intensity.
Is paid AI search optimisation tooling worth the budget?
Above 100 tracked prompts per engine, yes. Otterly.AI, Profound, BrightEdge AI, and equivalents automate prompt running, citation detection, share-of-voice calculation, and competitive benchmarking at scale. Below 100 prompts, the manual tracking discipline is enough and forces a closer reading of the actual responses. The case study deliberately used no paid tool to demonstrate reproducibility on a 60-hour budget. For agencies running multiple client programmes simultaneously or in-house teams tracking 200+ prompts across 6+ engines, paid tools become operationally necessary rather than optional.