Generative engine optimisation is the work of making a brand the source an AI answer reaches for, rather than one of ten blue links a user may never scroll to. The mechanics are less mysterious than the acronyms suggest: every major engine draws on a crawled index, every engine publishes the crawler that decides whether a site can be cited, and Google has now said in its own documentation that optimising for generative AI search is still SEO. What changes is which pages qualify, what earns a citation inside a synthesised answer, and how any of it can be measured.
Key Highlights
- Eligibility is the gate: Google states that a page can appear as a supporting link in AI Overviews and AI Mode if it is indexed and eligible to be shown with a snippet, and that “there are no additional technical requirements”.
- Each engine has its own crawler: OAI-SearchBot decides whether a site can be surfaced in ChatGPT’s search answers, GPTBot governs training, and Googlebot governs Google’s AI surfaces, so access has to be granted deliberately rather than assumed.
- Google names GEO and AEO and folds both into SEO: its May 2026 guidance describes AI features as rooted in core Search ranking and quality systems.
- Several popular tactics are explicitly unnecessary: Google says no AI text files, no special schema, no content chunking and no AI-specific rewriting are needed.
- Measurement already exists: AI Overviews and AI Mode performance appears in Search Console under the Web search type, and server logs show which AI crawlers actually arrive.
What Is Generative Engine Optimisation?
Generative engine optimisation is the practice of improving how often and how prominently a source is used in answers produced by systems that synthesise text from retrieved documents. The term comes from academic work rather than from the agency world. The paper “GEO: Generative Engine Optimization” by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande introduced both the name and GEO-bench, a benchmark of queries paired with the sources that could answer them, and reported visibility gains of up to 40 per cent from content changes, with effectiveness varying sharply by domain.
Two neighbouring terms circulate alongside it. Answer engine optimisation describes the same goal framed around direct answers, including featured snippets and voice results, and predates the current wave. LLM SEO is used loosely for the whole field. None of the three is a separate technical discipline with its own ranking system, which is the point Google made when it published guidance on optimising for generative AI features on 15 May 2026: it defines both answer engine optimisation and generative engine optimisation, then states that optimising for generative AI search is optimising for the search experience, and so remains SEO.
That position deserves to be read carefully rather than dismissed as self-interest. Google’s AI surfaces are built on the Search index and its ranking systems, so a page that cannot be crawled, indexed or shown with a snippet has no route into an AI Overview regardless of how its sentences are arranged. The practical implication for an Indian brand is that generative engine optimisation starts as an eligibility and quality exercise, not as a new channel.
How Do ChatGPT, Gemini and AI Overviews Choose What to Cite?
Retrieval comes first in every case. The engine turns a prompt into one or more searches, collects candidate documents, and then writes an answer that references some of them. Google describes this expansion explicitly as query fan-out, where its systems issue several related searches and surface a wider and more diverse set of links than a single query would produce. One consequence is that a brand can be cited for a prompt it never targeted, because the fan-out touched a sub-question the brand happens to answer well.
Access is governed separately by each provider, and conflating the crawlers is the most common technical error in this area.
| Surface | Crawler that decides citation | Control the site owner holds | Where to verify |
| Google AI Overviews and AI Mode | Googlebot | robots.txt rules for Googlebot, indexing, and snippet settings | Search Console URL Inspection and the Performance report |
| Gemini grounding and Google model training | Google-Extended | the Google-Extended token in robots.txt | server logs |
| ChatGPT search answers | OAI-SearchBot | allow or disallow OAI-SearchBot | server logs, with changes taking about 24 hours |
| ChatGPT model training | GPTBot | disallow GPTBot without affecting search eligibility | server logs |
| ChatGPT user-initiated visits | ChatGPT-User | robots.txt rules may not apply, and this agent does not decide search appearance | server logs |
Two details in that table are worth stating plainly. OpenAI’s documentation says a site opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though it can still appear as a navigational link, and that blocking GPTBot to prevent training use does not remove a site from search. On the Google side, Google-Extended covers training and grounding in Google’s other systems, so blocking it does not withdraw a site from AI Overviews or AI Mode, because those are Search surfaces governed by Googlebot and by snippet settings. Many brands have blocked the wrong token and concluded that AI search ignores them.
GEO vs SEO: What Actually Changes?
The honest answer to the GEO vs SEO question, and to most of what is sold as LLM SEO, is that the inputs barely change while the unit of competition does. Classic search competes for a position on a results page. A generative answer competes for a sentence inside a paragraph, often attributed to two or three sources, sometimes drawn from a page that ranks fifth or eighth for the query a user never typed.
Three differences follow from that. Specificity beats breadth, because a synthesised answer pulls the passage that resolves a narrow question rather than the page that covers a topic broadly. Extractability matters, which means important claims sit in text rather than in images, carousels or scripts, with semantic HTML and clean rendering, all of which Google lists among its foundational recommendations. And consistency across the web matters more than it did, because an engine reconciling conflicting figures about a company will often cite neither.
The basics haven’t changed. Crawlability, indexing, page experience, internal linking, less duplicate content and structured data that matches the visible text still matter, just as they did in traditional search. Google’s advice on what to avoid is worth noting too. There’s no need to create machine-readable AI files like llms.txt, break content into smaller fragments just for AI, rewrite pages for machines, add AI-specific schema or chase fake mentions. Google points out that these mentions may not be as useful as people expect.
Which Technical Settings Decide Whether You Can Be Cited?
Eligibility has three layers, and all three are checkable in an afternoon. The page must be crawlable, which means no robots.txt block and no CDN or host rule that quietly rejects the crawler. It must be indexed, which URL Inspection in Search Console will confirm. And it must be eligible for a snippet, because Google ties AI feature appearance to snippet eligibility. A nosnippet directive, a data-nosnippet attribute around the key passage, or a restrictive max-snippet value will each suppress the text an AI feature would otherwise use, and sites often carry these from an old decision nobody revisited.
Rendering is the next layer. Content that only appears after client-side execution is at the mercy of how a given crawler renders, and AI Overviews optimisation fails quietly when the answer a brand wants quoted is injected by a script. Server-rendered text, headings that describe the content beneath them and tables marked up as tables all help. Structured data remains useful for rich results eligibility, and Google is explicit that no AI-specific schema exists and that markup should match what a visitor can see.
Changes don’t show up straight away. Google says recrawling can take anywhere from a few days to several months, while OpenAI’s search refresh takes around 24 hours. So, if you fix eligibility issues in the first week, don’t expect to see clear results by the second week.
What Content Earns a Citation Once You Are Eligible?
Google’s strongest content instruction in the May 2026 guidance is to create non-commodity content, meaning material that offers insight beyond common knowledge. That is a useful filter for Indian brands publishing in crowded categories, because a page that restates what twenty other pages say gives a generative engine no reason to prefer it, while a page carrying original data, a method, a named source or a specific local condition gives it something only that page can supply.
Four simple content habits make sense when you look at how retrieval works. Answer the main question right at the start of the relevant section, since that’s the part most likely to be picked up. When using figures, mention the source and date so the information is easier to verify. Keep entity details consistent across your website, Business Profile and directories, because different addresses, founding years or team sizes can make it harder for an engine to confidently understand and summarise the brand. Also, cover the related questions that a fan-out search might bring up instead of spending too much time guessing keywords.
The academic finding supports the same direction. The GEO work reported that adding citations, quotations and statistics to content improved visibility in generative answers, while effects differed by domain, so the method generalises better than any single tactic does.
How Should an Indian Brand Phase This Over 90 Days?
- Weeks 1 to 2, eligibility audit: First, check that the important pages are indexed and eligible to appear in snippets. Review robots.txt for Googlebot, Google-Extended, OAI-SearchBot and GPTBot separately, check how key passages are being rendered, and remove any nosnippet or max-snippet settings that no one can properly justify.
- Weeks 3 to 6, content depth: Improve the twenty pages closest to commercial intent by building them around specific questions. Add original data or credible named sources, keep entity details consistent across all properties, and check that the structured data matches what is actually visible on the page.
- Weeks 7 to 10, measurement baseline: Use the same set of prompts across ChatGPT, Gemini and Google AI Mode, recording the date each time. In Search Console, look at the Web search type, and tag traffic coming from AI assistants in analytics so even the smaller amount of referral traffic can be tracked.
- Weeks 11 to 13, review and iterate: Run the same prompts again and compare the results with your baseline. Check which pages gained or lost citations, then use those findings to guide the next content plan instead of jumping straight into another set of tactics.
The order matters because each stage depends on the one before it. There’s little point improving content on pages that can’t be crawled, and starting measurement before those content changes have been recrawled can make everything look flat and lead to the wrong conclusions.
How Do You Measure AI Citations Honestly?
There are three main sources of truth, but none gives you the full picture. Search Console includes AI Overviews and AI Mode activity under the Web search type instead of showing them separately, so any growth you see there is more of a general direction than proof of performance on one specific surface. Server logs tell you which AI crawlers are actually requesting your pages, making them the most reliable way to confirm whether an access change has worked. Analytics referral data shows people who click through from an AI assistant, but this can be heavily undercounted because most people simply read the generated answer without clicking through.
AI Overviews optimisation needs to be tracked over several weeks, not judged by one screenshot. Prompt panels can help, though they don’t give the full picture. A team can test the same set of prompts regularly and note whether the brand appears and which page gets cited. This makes it easier to spot changes over time, as long as the prompts stay the same and each check is dated. Results can differ from one session or user to another, so a single check doesn’t tell you much. Looking at the results month by month gives you a better idea of what’s happening.
There are also two common mistakes to watch out for. No citations don’t automatically mean something is technically wrong, especially since Google has noted that AI Overviews don’t appear for every search. It’s also misleading to compare citation counts across different AI engines as if they were search rankings. Each engine uses its own retrieval process, model and refresh schedule, so the results won’t necessarily match.
Frequently Asked Questions
What is generative engine optimisation?
It is the practice of improving how often a brand’s content is retrieved and cited in answers generated by AI search systems. The term originated in the research paper that introduced GEO-bench, and it covers both eligibility work, such as crawling and snippet settings, and content work, such as answering narrow questions with verifiable detail.
Is generative engine optimisation different from SEO?
Google’s position, published in May 2026, is that optimising for generative AI features is optimising for the search experience and therefore still SEO, because its AI features draw on the same index and ranking systems. The practical difference lies in competing for a sentence inside an answer rather than a position on a page, which rewards specificity and extractable text.
How do I let ChatGPT cite my website?
Allow OAI-SearchBot in robots.txt, since OpenAI states that a site opted out of it will not be shown in ChatGPT search answers. Blocking GPTBot, which governs training, does not remove a site from ChatGPT search, so the two decisions can be taken separately. Allow about 24 hours for a change to register.
Does blocking Google-Extended remove my site from AI Overviews?
No. Google-Extended governs training and grounding in Google’s other systems, while AI Overviews and AI Mode are Search surfaces controlled through Googlebot and your snippet settings. Sites that block Googlebot or disable snippets are the ones that lose eligibility.
Do I need an llms.txt file to be cited?
No. Google’s documentation states that no new machine-readable files, AI text files or special markup are needed, and Google representatives have said AI systems are not requesting the file. Effort is better spent on crawlability, indexing and content that says something the rest of the category does not.
How long does it take to appear in AI answers?
Eligibility changes can register within days, but Google notes recrawling may take from several days to several months, so a realistic review point is 8 to 12 weeks after the work is done. Content depth usually takes longer to show than a technical fix does.
What Should Brands Fix First?
Getting cited is a disciplined version of work most brands already do badly: make the page reachable, make it eligible for a snippet, say something that is not available in twenty other places, and keep the facts about your business identical everywhere they appear. The acronyms will keep changing, and the eligibility checklist will not.
At Savit Interactive, we have been doing search work from Mumbai since 2004, and our AI SEO services in Mumbai sit alongside the AI SEO Playbook we publish rather than replacing the fundamentals our team has always run on. We treat answer engine optimisation and generative engine optimisation as extensions of a single programme, which is why our organic search teams audit crawler access, snippet eligibility and entity consistency before anyone touches a content calendar. With 100+ specialists and work delivered for 5000+ clients across apparel, B2B, healthcare and pharma, education, ecommerce, logistics, manufacturing and AI and LLM businesses, we see the same pattern repeatedly: the brands that get cited are the ones that fixed eligibility first and then published something worth quoting.
