AIO Guide 2026: get cited by ChatGPT and Google

By Million MarketingReading time: 11 min

On 22 July 2026, Google switched on AI Overviews in its search engine in France [3]. For brands, the consequence is immediate: on a growing share of queries, the first thing users see is no longer a list of links but an answer written by an AI, which cites — or does not cite — its sources. In the United States, 49% of adults now use a chatbot such as ChatGPT, Gemini or Copilot, and 60% say they read the AI-generated summaries at the top of search results [1].

AIO (Artificial Intelligence Optimization) covers all the practices that make a brand, its content and its data readable, credible and citable by these answer engines. This guide brings together what can be verified in 2026: figures on the impact on traffic, the official positions of Google and OpenAI, the findings of academic research, measurement tools and the European legal framework. Every figure links to its source, listed at the end of the article.

AIO defined: from ranked page to cited source

Classic search engine optimisation aims for a position in a list of results. AIO aims for a citation in an answer. The difference is structural: assistants such as ChatGPT Search, Perplexity, Gemini or Google’s AI Overviews mostly work on retrieval-augmented generation (RAG). The system queries an index, selects a few passages it deems relevant, then synthesises them into an answer, citing some or all of the pages used.

Three conditions must therefore be met for a brand to appear: its pages must be accessible to the crawlers that feed these indexes, its passages must be selected from dozens of candidates, and its content must be judged reliable enough to be reused and attributed. The terms “AIO” and “AIO” are also used for the same discipline; we use AIO here, a broader term covering both generative search engines and conversational assistants.

AIO does not replace SEO: it extends it. Google says so explicitly — search fundamentals remain the foundation of visibility in its AI features [8][9]. Our SEO approach and our AIO offer are designed as a continuum.

Why 2026 changes the game: what the figures say

Studies published since 2024 agree on one point: when an AI-generated answer appears, users click less on classic results. The scale varies with the methodology, which is why they are worth reading together.

Study	Scope	Key finding
Pew Research Center (July 2025) [2]	900 US adults, 68,879 searches observed	Click on a classic result in 8% of visits with an AI summary, versus 15% without; only 1% click a link inside the summary
Seer Interactive (November 2025) [4]	Queries with AI Overviews, June 2024 to Q3 2025	Organic CTR −61% (from 1.76% to 0.61%); paid CTR −68%
Ahrefs (February 2026, updated August 2026) [6]	300,000 keywords, December 2025 data	Position-1 CTR reduced by 58% when an AI Overview is present
SparkToro / Datos (July 2024) [7]	Google searches in the United States and the European Union	58.5% (US) and 59.7% (EU) of searches end without a click

Two nuances are worth stressing. First, being cited changes everything: according to Seer Interactive, brands cited in an AI Overview earn 35% higher organic CTR and 91% higher paid CTR than those that are not [4]. Second, the trend is not linear: the April 2026 update of the same study observes a rebound in organic CTR on queries with AI Overviews, from 1.3% in December 2025 to 2.4% in February 2026 [5]. The challenge is therefore not to mourn the end of the click, but to be among the sources selected.

For France, the activation of AI Overviews in summer 2026 [3] means that these dynamics, observed in the United States for two years, now apply to French-language queries. Advertisers who also run paid campaigns should revisit their CTR assumptions: our Google Ads budget calculator lets you simulate the effect of a lower click-through rate on cost per acquisition.

What Google says (and does not say) about optimising for AI

Google Search Central’s documentation is unambiguous: there are no additional requirements to appear in AI Overviews or AI Mode, nor any special optimisation needed [8]. A page must be indexed and eligible to be shown with a snippet in order to be used. The AI optimisation guide published by Google, updated in July 2026, clarifies several points that are often misunderstood [9]:

SEO remains the foundation. The same ranking and quality systems power the AI features.
No specific file is required for Google. Neither an llms.txt file nor AI-specific markup determines eligibility in Google Search.
Structured data is not mandatory for AI features, although it remains useful for rich results.
Artificially chunking or rewriting content “for AI” does not help and may breach spam policies.

Publishers keep control over how their content is reused through the usual directives: nosnippet, data-nosnippet, max-snippet or noindex [8][18]. Restricting the snippet also mechanically limits the chance of being cited.

Finally, spam policies apply in full. Google defines scaled content abuse as generating many pages primarily to manipulate rankings rather than to help users, whatever the production tool [11]. Its site reputation policy, revised in August 2026, targets third-party content hosted to take advantage of the host site’s ranking signals [11]. Producing hundreds of AI-generated pages to “saturate” answers is therefore a risky strategy.

Technical foundations: being readable by AI crawlers

Telling OpenAI’s crawlers apart

OpenAI documents several distinct agents that can be allowed or blocked separately in robots.txt [12]:

Agent	Role	Effect of blocking
GPTBot	Collects content to train the models	Content is not used to train future models
OAI-SearchBot	Indexing for ChatGPT search results	The site does not appear in ChatGPT Search answers
ChatGPT-User	Visits a page at a user’s request	robots.txt rules may not apply to these requests

A brand can therefore opt out of training (GPTBot) while staying visible in ChatGPT search (OAI-SearchBot). This is often the most relevant trade-off for a commercial site: visibility in answers is an acquisition issue, training is an intellectual property issue.

The JavaScript trap

An analysis by Vercel of its network estimated at the end of 2024 that GPTBot made 569 million requests a month and Claude’s crawler 370 million, about 20% of Googlebot’s volume [13]. Above all, it found that none of the major AI crawlers then executed JavaScript, with the exception of Gemini (via Googlebot’s infrastructure) and AppleBot [13]. A site whose content only appears after scripts run in the browser may therefore be partly invisible to these engines. Server-side rendering or static generation of essential content (text, reference prices, FAQs, product information) is a basic precaution.

Structuring for understanding

Even though Google does not require it for its AI features, consistent structured data (Organization, Service, Product, FAQ, Article) helps engines link a page to a clearly identified entity. Consistency of brand information across the web — website, business listings, professional directories, press — plays the same role: an answer engine is more likely to cite an entity whose attributes match from one source to another.

Editorial levers that increase the likelihood of being cited

The academic reference on the subject is the “Generative Engine Optimization” study by researchers from Princeton and IIT Delhi, presented at KDD 2024 [14]. On a benchmark of 10,000 queries, the authors measured how different content changes affected a source’s visibility in generated answers. The most effective methods increased that visibility by up to 40%:

Adding quotations from experts or recognised sources: +27.2%.
Adding precise, sourced statistics: +25.2%.
Improving fluency of the writing: +24.7%.
Citing sources: +24.6%.
Conversely, keyword stuffing reduced visibility (−8.2%).

These results were obtained on a controlled benchmark rather than on each commercial engine; they nonetheless point to a trend consistent with Google’s quality guidelines: factual, precise and sourced content is what answer systems favour.

A second research finding sheds light on page structure. The “Lost in the Middle” study (Stanford, 2023) showed that language models make better use of information at the beginning or end of a long context than in the middle [15]. Without mechanically transposing this to commercial engines, it argues for writing that gives the answer upfront: a clear definition at the top of each section, then the detail.

In practice, we recommend five editorial rules:

Answer first. Each section opens with a sentence that answers the question in its heading.
Quantify and source. A dated, attributed and linked data point beats a general claim.
Write for real questions. Queries put to assistants are longer and more conversational; headings and FAQs should reflect them.
Show expertise. Named authors, real client references, verifiable case studies: all signals of reliability.
Keep content up to date. Answer engines favour recent information on fast-moving topics; a visible and genuine update date matters.

Measuring AIO: from intuition to metrics

For a long time, visibility in generated answers was hard to measure. Three tools now make it possible to track it.

Google Search Console. In June 2026 Google announced new performance reports dedicated to its generative AI features, rolled out gradually [10]. They make it possible to track a site’s presence in AI Overviews and AI Mode separately from classic results.

Google Analytics 4. A custom channel group isolates traffic from assistants. Simply create an “AI assistants” channel placed above the “Referral” channel, with a session source condition matching a regular expression such as chatgpt\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai. These visits are often few in number but highly qualified, as users arrive after already comparing the options.

Generative share of voice. The most strategic indicator remains manual or semi-automated: define a panel of 30 to 100 questions representative of the buying journey, submit them each month to ChatGPT, Gemini, Perplexity and AI Overviews, then measure how often the brand is cited, recommended or absent compared with its competitors. It is the AI equivalent of rank tracking in SEO.

The legal framework: what to know in Europe

Article 50 of the European Artificial Intelligence Regulation (AI Act, Regulation (EU) 2024/1689) has applied since 2 August 2026 [16]. It imposes transparency obligations on providers and deployers of AI systems, including the labelling of synthetic content (images, audio, video, deepfakes and certain texts published to inform the public). Brands that use generative AI in their content production must build these obligations into their editorial processes. We cover them in detail in our article on the AI Act and Article 50.

On the publishers’ side, the relationship with AI players is becoming contractual. In March 2024, Le Monde signed a partnership with OpenAI covering the use of its content and its promotion, with links, in ChatGPT’s answers [17]. For a brand, the question is not to sign such agreements, but to make a conscious trade-off, via robots.txt and snippet directives, between protecting content and being visible in answers.

A four-step AIO roadmap

Diagnosis (weeks 1 to 3). Build the question panel, measure the current generative share of voice against three to five competitors, audit robots.txt, page rendering without JavaScript and the consistency of brand information.
Technical foundations (weeks 3 to 6). Decide on AI crawler access, ensure server-side rendering of key content, harmonise structured data and entity information across the web.
Reference content (weeks 6 to 12). Prioritise the pages that answer the panel’s questions: service pages, pillar guides, FAQs, case studies. Enrich them with sourced data, quotations and direct answers.
Measurement and iteration (ongoing). Track generative share of voice, Search Console AI reports and the “AI assistants” channel in GA4 every month, then shift editorial effort towards the questions where the brand is still absent.

Our approach at Million Marketing

For more than 20 years, we have supported large accounts and mid-market companies with their digital acquisition, from SEO to media campaigns. Our AIO approach follows on from this: it starts with an objective measurement of the brand’s presence in AI answers, builds on solid SEO foundations and favours factual, sourced content that complies with Google’s rules and European law. No mass production, no promise of guaranteed citations: a method, metrics and long-term management.

Discover our AIO offer →

Frequently asked questions about AIO

What is the difference between SEO and AIO?

SEO aims to rank a page in a list of results; AIO aims to get a brand or a piece of content cited in an answer generated by an AI. Both rest on the same foundations — accessibility, quality, authority — but AIO adds requirements for factual accuracy, structure and measurement of share of voice in assistants.

Should you create an llms.txt file?

Google states that no specific file, including llms.txt, is needed to appear in its AI features [9]. This file can serve as a readable summary for some assistants, but it does not replace quality content or good indexing.

Should you block GPTBot in robots.txt?

Blocking GPTBot prevents your content from being used to train OpenAI’s models, without excluding the site from ChatGPT search, which relies on OAI-SearchBot [12]. The decision is a trade-off between content protection and visibility; it should be made agent by agent.

How can I tell whether my brand is cited by ChatGPT or Gemini?

By combining three sources: Search Console’s generative AI reports for Google [10], a dedicated assistants channel in GA4 for incoming traffic, and monthly tracking of a panel of questions submitted to the main assistants to measure share of voice against competitors.

For a first benchmark, our AIO Barometer 2026 measures the brands cited by ChatGPT, Gemini and Google across 40 buying questions in eight sectors.

AIO Guide 2026: how to get cited by ChatGPT, Gemini and AI Overviews