AI SEO: How to Rank in ChatGPT, Gemini and Claude

Learn how to rank in ChatGPT, Gemini, Claude, Perplexity and Copilot using proven AI SEO strategies

A growing share of the research your buyers do now happens inside a chat window, not a search results page. When a Marketing Director at a mid-market financial services firm asks ChatGPT to shortlist growth partners, or a CTO asks Gemini to explain a category before a vendor call, the businesses that get named in that answer have already won half the sale — before a single click happens.

This is the part most companies are still getting wrong: they’re optimising for a results page fewer buyers scroll through, while the AI systems now sitting between the buyer and the search engine quietly decide who gets considered.

This guide sets out what’s actually verified — from Google, OpenAI, Microsoft and peer-reviewed research — about how these systems choose what to cite, and what a B2B business in the R20M–R100M revenue band should do about it.

What “AI SEO” actually means

The industry has settled on a few overlapping terms — AI SEO, Generative Engine Optimization (GEO), Answer Engine Optimization (AEO) — that all describe the same shift: optimising content so generative AI systems (ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google’s AI Overviews) surface it as a trusted answer, rather than optimising purely for a ranked position on a results page.

The distinction that matters commercially: traditional SEO competes for a click. AI SEO competes for a citation — being the passage a model extracts, paraphrases, and attributes when it answers a question. A buyer researching vendors inside an AI chat may never see a results page at all; the only way into that conversation is to already be the source the model trusts.

Academic researchers first formalised this in 2023, when a team from Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi published “GEO: Generative Engine Optimization” — the paper that coined the term. They built a 10,000-query benchmark named GEO-BENCH, spanning nine domains and tested which content interventions measurably increased a source’s visibility inside generative answers. Techniques such as adding cited statistics and authoritative-sounding language produced visibility gains of roughly 30–40% on their benchmark, with no change to the underlying facts of the content — a signal that how information is structured and supported matters as much as the information itself.

Why this matters now: the scale shift

The volume of queries now being answered inside AI systems, rather than on a traditional results page, has crossed a threshold that makes this a commercial issue rather than a technical curiosity.

Note on the Perplexity figures: reported numbers vary significantly across sources depending on whether they count core app users, web visitors, or API-driven integrations. Treat any single figure with caution and prioritise the direction of travel — sustained growth — over the precise number.

The pattern across all four platforms is the same: this is no longer a niche channel. For a business whose buying committee sits at CEO, CFO or CMO level — smaller, more senior, and more likely to use AI tools directly for pre-purchase research — the exposure is proportionally higher than it is for a mass-consumer brand.

Not sure whether your business currently shows up in those answers? Get in touch and we’ll run the check with you.

What Google, OpenAI and Microsoft actually say

There is a lot of noise in this space from agencies selling “GEO packages.” It’s worth separating that from what the platforms themselves have actually published.

Google. Google’s own developer documentation on AI features (published via Search Central) is explicit that there is no separate optimisation framework required for AI Overviews or AI Mode: no special markup, no new machine-readable files, no dedicated schema type. Instead, Google states that existing SEO fundamentals — crawlability, indexability, internal linking, and content quality — remain the basis for appearing in AI-generated answers, and that content already performing well organically has a higher baseline probability of being surfaced. This guidance is consistent with Google’s long-standing E-E-A-T framework (Experience, Expertise, Authoritativeness, Trust), which its Search Quality Rater Guidelines use to describe what “good” content looks like.

OpenAI. OpenAI has published less prescriptive guidance for publishers, but its help documentation on ChatGPT Search describes the underlying mechanism: search-enabled responses use a retrieval-augmented generation approach, querying the web (via its own index and, for some query types, Bing-powered results) before generating an answer, then attaching inline citations a user can click through to the source. OpenAI has not published an official ranking or weighting methodology for which sources get selected in that retrieval step — a gap that has left room for third-party speculation, some of it more reliable than others.

Microsoft. Microsoft has moved furthest toward direct publisher tooling: in February 2026 it introduced “AI Performance” reporting inside Bing Webmaster Tools, giving site owners visibility into how often their content is cited across Copilot and Bing’s AI-generated summaries. Microsoft’s guidance is that Copilot grounds most web-sourced answers through the Bing search service, so visibility in Copilot starts with proper indexing and readability in Bing — verified site ownership, IndexNow enabled for fast content-change discovery, and clean meta robots directives. Microsoft’s own IndexNow guidance is specific: submit URLs when content actually changes, not on a fixed schedule, since repeated submission of unchanged pages “provides no value and may be interpreted as spam.”

The consistent thread across all three: none of them describe AI visibility as a separate discipline you bolt onto a site. They describe it as an extension of doing SEO properly, with a few added layers.

E-E-A-T, extended

Google’s E-E-A-T framework — Experience, Expertise, Authoritativeness, Trustworthiness — was written for human search quality raters evaluating traditional results. It has taken on new weight in an AI context for a simple reason: a model deciding what to cite is making a live trust judgement about a source, in real time, with no human reviewer in the loop. The signals that make content trustworthy to a human rater are largely the same signals a retrieval system uses to decide what’s safe to paraphrase and attribute.

For a financial services, telecoms or technology brand, this has practical implications:

• Experience — first-hand operational detail (methodologies, named case outcomes, specific numbers) reads as more citable than generic claims.
• Expertise — author attribution and credentialed bylines matter more than they used to, particularly for regulated or technical sectors where “who wrote this” is itself a trust signal.
• Authoritativeness — consistent, corroborated information about a company across multiple independent sources (not just the company’s own site) strengthens how models weigh a claim.
• Trustworthiness — accuracy, transparency about data sources, and up-to-date content reduce the risk a model treats a page as unreliable.

The framework: four things that determine whether you get cited

1. Direct answers, early. Every source reviewed here — Google’s own documentation, the academic GEO research, and platform-specific guidance — converges on the same structural point: content that states a clear position or definition in the first few sentences, then supports it with evidence, outperforms content that builds up to the point. Retrieval systems extract passages, not entire pages; a passage that already reads like a self-contained answer is easier to lift cleanly.

2. Evidence, not assertion. The Princeton/Georgia Tech research specifically tested this: adding statistics and citations to back up claims was one of the highest-impact interventions in their benchmark. “We deliver measurable growth” is filler. “We took a telecoms client from X to Y qualified leads per quarter in six months” is citable — it gives a model something concrete to extract and attribute.

3. Structured data — used correctly, not everywhere. Google is explicit that no special schema is required for AI Overviews. That doesn’t make structured data worthless — FAQPage schema in JSON-LD format remains one of the most useful formats because it packages content into the question-and-answer shape that mirrors how people actually prompt AI systems, and it’s a well-supported, low-risk addition to a technically sound site. It’s a refinement on top of fundamentals, not a substitute for them.

4. Consistency across the web. AI systems corroborate claims across multiple sources — reviews, press coverage, directories, LinkedIn — not just a company’s own site. Fragmented or contradictory information (different positioning, different numbers, different claims across channels) weakens the authoritativeness signal a model relies on. This is a data-hygiene problem before it’s a content problem.

Platform notes

What this looks like by sector

Financial services. Regulated categories carry an inbuilt trust deficit that AI systems seem to reflect, not overcome — vague claims about returns, compliance or risk management are exactly the kind of unsupported assertion the research suggests gets filtered out in favour of a competitor who cites a specific framework, regulator, or audited result. A page that names the actual compliance standard it operates under is more citable than one that gestures at “trusted, compliant service.”

Telecoms. Buyers in this category are often comparing technical specifications and SLAs against a small set of known providers — which makes comparison content (honest, specific, naming the trade-offs) unusually valuable. A model asked to compare providers has to extract that comparison from somewhere; a source that already lays out the trade-offs clearly, rather than only self-promoting, is easier to lift and attribute.

Technology. This sector has the most AI-literate buyers of the three, and the most content competing for the same queries. The differentiator here tends to be depth and specificity — implementation detail, named methodologies, real numbers — over broad thought-leadership pieces that say the same thing as every competitor’s blog.

Common mistakes

The most common failure mode is treating this as a checklist to bolt onto an otherwise weak site — adding FAQ schema to thin, unsubstantiated content, or chasing platform-specific tricks instead of the underlying signals every platform’s own guidance points back to. A second, subtler mistake is inconsistency: publishing a strong case study on the website while directories, LinkedIn and press mentions still carry outdated positioning or numbers. AI systems corroborate across sources; internal inconsistency is treated as a trust problem, not a branding oversight.

A third mistake is measuring the wrong thing. Teams that track only organic rankings will miss this shift entirely — a page can hold its ranking position while losing the citation inside the AI answer that used to sit above it. And a fourth: assuming this is a one-off project. Content that was well-structured a year ago can lose citation share as competitors catch up on the same fundamentals; this needs to sit inside an ongoing content operation, not a single audit.

How to measure this without guessing

First-party measurement tools are still catching up to the shift. Bing Webmaster Tools’ AI Performance report (introduced February 2026) is currently the most direct first-party visibility Microsoft or Google offer into AI citations. Neither OpenAI, Google’s AI Overviews, nor Perplexity currently publish an equivalent self-serve report for individual site owners.

In the absence of that, a workable interim approach: build a fixed list of 15–20 branded and category queries a real buyer would plausibly ask, run them manually across ChatGPT, Gemini, Perplexity and Copilot on a monthly cadence, and log whether the business is mentioned, cited with a link, or absent. It’s manual, but it produces a real baseline and a trend line — which is more than most competitors are currently tracking.

A 90-day approach

For a business in the R20M–R100M band, this doesn’t require a rebuilt content operation. A phased approach:

Days 1–30 — Foundation audit. Confirm technical SEO is actually sound (not assumed sound): crawlability, indexing, site speed, mobile rendering. Audit consistency of company information — name, positioning, numbers, claims — across the website, LinkedIn, directories and press.

Days 31–60 — Content restructuring. Rewrite priority pages (service pages, comparison content, cornerstone guides) to lead with direct, evidenced answers rather than narrative build-up. Add FAQPage schema to genuinely useful Q&A content, not as a bolt-on.

Days 61–90 — Authority building and measurement. Pursue corroborating third-party coverage (press, directories, partner mentions) to strengthen cross-source consistency. Set up Bing Webmaster Tools’ AI Performance reporting and monitor citation appearances where available; track branded and category queries inside AI tools manually as a baseline, since third-party AI-rank tracking is still maturing.

FAQs

Is GEO replacing SEO? No. Every primary source reviewed here — Google’s documentation, Microsoft’s guidance, and the academic literature — describes AI visibility as built on top of traditional SEO fundamentals, not as a separate discipline that replaces them.

Do I need special schema markup to appear in AI Overviews? Google’s own guidance states no special schema is required. FAQPage schema remains useful because it structures content in a way that mirrors how people prompt AI systems, but it’s a refinement, not a requirement.

Which platform should a B2B company prioritise first? Start with the one your buyers actually use for research. For most financial services, telecoms and technology buyers in the R20M–R100M band, that’s typically ChatGPT and Google’s AI features first, with Copilot relevant if the buyer’s organisation is Microsoft-centric.

How do I know if I’m being cited? Bing Webmaster Tools now reports AI citation visibility directly. For ChatGPT, Gemini and Perplexity, there is no equivalent first-party reporting yet — tracking currently means manually testing branded and category queries and logging results.

Does this apply outside financial services, telecoms and technology? The underlying mechanics are sector-agnostic, but the impact is proportionally larger for sectors with senior, research-heavy buying committees — which is why it matters more here than in high-volume consumer categories.

References

• Google Search Central — AI Features and Your Website
• Bing Webmaster Blog — Introducing AI Performance in Bing Webmaster Tools (Public Preview)
• OpenAI Help Center — ChatGPT Search
• OpenAI — Introducing ChatGPT Search
• Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande — GEO: Generative Engine Optimization, Princeton University / Georgia Tech / Allen Institute for AI / IIT Delhi, presented at ACM SIGKDD 2024
• Reuters / Sensor Tower — ChatGPT monthly active app user milestone reporting, June 2026
• Google Q2 2026 disclosure — AI Overviews monthly user figures