First published: 4 May 2026 · Last updated: 4 May 2026
ChatGPT Search
- Heavy weight on recency
- Prefers structured pages
- Bing visibility = ChatGPT visibility
Perplexity
- Strong recency bias on news topics
- Cites 4-6 sources per answer
- Easy-to-extract passages win
Google AI Overviews
- E-E-A-T heavy
- Schema markup matters
- Topical authority compounds
Claude (with Search)
- Prefers reputable publications
- Smaller citation set per answer
- Authority > freshness
Bing Copilot
- Functionally similar to ChatGPT
- Stronger Microsoft ecosystem bias
- Often skipped in SG audits
The Binary Nature of AI Search
Classical SEO has positions. Position 1 captures more traffic than position 2, which captures more than position 3, and so on down the SERP. The whole game is incremental ranking improvement. AI search does not work that way. There is no position 2. Either your content is cited inside the AI answer (and named in the source list) or it is invisible. The user reads the answer, sometimes clicks one of the cited sources, and rarely scrolls. The middle ground that classical SEO operated in (positions 4 to 10) does not exist in AI search. This binary structure changes optimisation priorities. Marginal ranking improvements that move you from position 7 to position 5 in classical SERPs do nothing for AI search. The only meaningful work is moving from "uncited" to "cited", which is a different threshold and requires different signals.The 4 Cross-Engine Signals That Lift Citations
Independent studies and our own client work converge on the same four signals. Each lifts citation rates by a measurable amount across all 5 engines.These four are the foundation. Add them to every page targeting an AI-eligible query before you do any engine-specific work. The compounding effect is significant: a page with all four signals is roughly 2 to 3 times more likely to be cited than a page with none, regardless of which AI engine the query runs on.
How to operationalise each signal:
- Cite sources inline. Hyperlink to authoritative external sources for any claim that is not common knowledge. AI engines treat this as a trust signal. Use `rel=""` (do not nofollow) for sources you genuinely vouch for.
- Add statistics with attribution. "Singapore had 4.7 million digital banking users in 2025 (MAS, Annual Report 2025)" wins over "many Singaporeans use digital banking". Numbers + named source = citable passage.
- Include expert quotes. A single quote from a named expert with a verifiable role lifts citations more than three paragraphs of general analysis. Pull from interviews, podcasts, or original outreach.
- Lead every H2 with a 40-60 word direct answer. Then expand. AI engines lift the opening passage of the relevant section, not the whole section.
For a deeper look at the topical structure that makes these signals compound, our content strategy service covers the cluster architecture that turns individual page optimisations into multi-engine domain authority.
How ChatGPT Search Actually Picks Sources
ChatGPT Search runs on the Bing index plus OpenAI's synthesis layer. Practically, this means:
Bing index retrieval
ChatGPT pulls candidate sources from the Bing search index. Pages that rank well in Bing classical search rank well in ChatGPT Search. Pages absent from Bing are absent from ChatGPT entirely.
Recency filter
For time-sensitive queries (news, technology, prices), ChatGPT applies a stronger recency weighting than Google AI Overviews. Pages updated within 30 days outperform older pages on these queries.
Synthesis + citation
OpenAI's model synthesises a 3-7 paragraph answer and embeds inline citations. Citations link back to source URLs. Cited URLs appear as numbered references at the bottom of the response.
ChatGPT-specific tactics:
- Get indexed in Bing first. Submit your sitemap via Bing Webmaster Tools. This is a frequent oversight in SG agency setups where Bing is treated as an afterthought.
- Refresh dates matter. Include `dateModified` in Article schema and visible "Last updated" text. ChatGPT's recency filter reads both.
- Structure for excerpt clarity. Short paragraphs (under 60 words), clear H2/H3 hierarchy, and bullet lists where natural. ChatGPT's lifter prefers atomic passages.
How Perplexity Actually Picks Sources
Perplexity runs its own crawler (PerplexityBot) and synthesises across 4 to 6 sources per answer. It cites more aggressively than other engines and updates its index more frequently.
PerplexityBot crawl
Perplexity maintains its own web index via PerplexityBot. Robots.txt rules apply. Sites that block PerplexityBot will not appear in Perplexity citations even if they rank well in Google.
Multi-source synthesis
Perplexity selects 4-6 sources per query, weights them by relevance and authority, and synthesises an answer that pulls from all of them. The Pro version (paid) accesses a wider candidate pool than the free version.
Source attribution
Each cited source appears as a numbered card in the answer interface. Cards include the page title, domain, and a short preview. Click-through rates from Perplexity citations are higher than from Google AI Overviews citations.
Perplexity-specific tactics:
- Allow PerplexityBot in robots.txt. Default Apache/Nginx configurations sometimes block it. Add `User-agent: PerplexityBot \n Allow: /` explicitly.
- Recency wins more here than other engines. For news, regulations, and tech queries, Perplexity heavily favours pages updated within 7 days. For evergreen topics, depth wins.
- Optimise for "card preview". Your title tag and meta description appear in the citation card. A boring meta description costs you click-through.
Google AI Overviews (Brief Recap, See Day 2 for Depth)
We covered Google AI Overviews in detail in our SGE to AI Overviews deep-dive. Headline points relevant to multi-engine optimisation:
- Powered by Google index + Gemini synthesis
- Heavier E-E-A-T weighting than other engines
- Schema markup (FAQPage, HowTo, Article) actively moves the needle
- Information Gain scoring penalises regurgitation
If you optimise for AI Overviews well, you usually capture some Bing/ChatGPT lift as a side effect (because both reward similar structural signals), but Perplexity and Claude need their own attention.
How Claude Picks Sources
Claude's web search (introduced in 2025, expanded through 2026) is more conservative than other engines. It retrieves a smaller candidate set and weighs source authority heavily.
Claude-specific patterns observed:
- Authoritative publications get cited disproportionately. Reuters, BBC, established trade publications, government sites. Lower-authority blogs are cited less even when content quality is comparable.
- Smaller citation sets per answer. Claude often cites 2-3 sources where Perplexity cites 4-6. Getting picked is harder.
- Cleaner structured pages preferred. Claude's retrieval seems to favour pages with clear hierarchy, valid schema, and minimal interstitials.
- Less recency emphasis. Claude weights authority over freshness more than ChatGPT or Perplexity.
Claude-specific tactics:
- Build E-E-A-T evidence aggressively. Author bios, credentials, original first-hand experience all matter more here.
- Pursue brand mentions in trusted publications. PR-style coverage in Reuters or trade publications has outsized impact on Claude visibility.
- Schema validation matters. Run pages through Schema.org validator and Google's Rich Results Test.
Bing Copilot (Brief)
Bing Copilot is functionally similar to ChatGPT Search (both use the Bing index + GPT models). Optimising for ChatGPT Search captures most of the Bing Copilot lift automatically. Worth a separate prompt-test cycle if you serve enterprise Microsoft-heavy clients, otherwise treat as a sub-engine of the ChatGPT optimisation work.
The 4-Engine Baseline Test (Run This First)
Before any engine-specific tactical work, baseline where you stand. Run this audit once per quarter.
Pick 10 buyer-intent queries
The exact phrases your prospects would type when researching solutions in your category. Mix of "what is", "how to", "best", and comparison queries.
Run each query in 4 engines
ChatGPT Search, Perplexity, Google AI Overviews, Claude (with web search). Same wording in all four. Capture screenshots of every answer.
Score per query: cited or invisible
For each of the 40 answers (10 queries × 4 engines), record whether your brand is cited, mentioned, paraphrased, or absent. Track which competitors are cited.
Identify engine-specific gaps
If you are cited well in Google AI Overviews but absent from Perplexity, you have a recency or PerplexityBot indexing problem. Diagnose per engine.
Re-run quarterly
Use the same 10 queries every quarter to measure delta. Add 2-3 new queries each quarter for emerging buyer language.
For a foundation in the underlying technical SEO setup that all 4 engines reward, this audit only diagnoses. Remediation is a separate workstream.
Frequently Asked Questions
Which AI engine should I optimise for first?
Google AI Overviews if you serve a Singapore consumer audience (highest reach). Perplexity if you serve a research-heavy B2B audience (higher conversion intent). ChatGPT Search if you depend on Bing-heavy enterprise traffic. Optimise for one engine to a working baseline, then expand. The cross-engine signals (sources, stats, quotes, answer-first structure) carry roughly 70% of the work across all engines.
How is ranking in AI engines different from traditional SEO?
AI engine ranking is binary: cited or invisible. There is no position 2. Classical SEO ranks pages incrementally on a SERP; AI engines either pull your passage into their answer or ignore you entirely. This changes the optimisation calculation: marginal classical ranking improvements (position 7 to 5) do not lift AI citations. Only crossing the "cited" threshold matters.
Does PerplexityBot crawl my site by default?
Yes if your robots.txt does not block it. Some default Nginx and Apache configurations include rules that accidentally block AI crawlers. Check your robots.txt explicitly: there should be no `Disallow` rule that targets PerplexityBot, GPTBot, ClaudeBot or CCBot. We cover the full AI crawler configuration in our upcoming GPTBot guide.
How much does ranking in AI engines actually lift traffic?
Direct click-through from AI citations is currently 1-5% of classical SERP click-through for comparable visibility. The bigger effect is brand mention impact: even uncited brand mentions inside AI answers move pipeline metrics (branded search volume, direct traffic, conversion rate). For decision-maker framing on this, see our sister article on bestmarketing.com.sg.
Are there tools to monitor AI engine rankings?
Yes. Profound (enterprise multi-engine), Athena HQ (focused on AI Overviews), and Otterly (lighter weight, agency-friendly) are the maturing options as of Q2 2026. Manual prompt testing on a stable query set still produces the highest signal-to-noise for most teams under SGD 5,000/month tooling budget.