AI research highlights the pages that demonstrate the highest quality, the strongest presence and the greatest clarity. This guide brings together what official sources and verified studies have to say. It then translates all of this into priorities.
There are two main approaches. Each has its own rules.
Google explains that these features use retrieval-augmented generation (RAG) and query fan-out to highlight content from the search index (Semrush). The index is the same as for traditional search. Being in the index is the prerequisite for appearing in AI responses.
These assistants use their own crawlers and indexes. The OAI-SearchBot builds the index for ChatGPT’s search (GEOToolbox). The Claude-SearchBot indexes pages for Claude’s search (The Rank Masters). The PerplexityBot performs the same function on Perplexity.
Three conditions recur across all the sources consulted:
These three concepts form part of the same digital visibility strategy. Each has its own focus and addresses different ways of searching for and discovering information.
| Layer | Objective | Measure of success |
|---|---|---|
| SEO | Position in search results | Rankings, impressions, clicks |
| AEO | Direct answers in snippets and AI Overviews | Presence in direct answers |
| GEO | Mentions and recommendations in AI responses | Mentions, citations, recommendations |
Source: WRITER
Google argues that optimisation for generative search experiences should be based on the same principles that guide traditional SEO. From this perspective, improving visibility in AI-generated results remains part of SEO work, as these experiences rely on the ranking and quality systems of conventional search (Google Search Central).
Nevertheless, the industry uses different terms to describe this development. AEO (Answer Engine Optimisation) focuses on optimising content for answer engines. GEO (Generative Engine Optimisation) seeks to improve brands’ visibility and presence in AI-generated answers. It is also common to come across the term ‘AI SEO’, which frames these practices as an evolution of traditional SEO.
On 15 May 2026, Google published the first official guide to optimising websites for AI-powered search features (Google Search Central). The key message is that good SEO practices remain the foundation.
Could an AI model write an equally useful version of this page? If the answer is yes, then something is missing that only the company has: its own data, project experience and informed opinion.
Source: Semrush and DemandSphere.
This guide applies to Google Search (AI Overviews and AI Mode). For ChatGPT, Claude and other assistants, the findings serve as best practice, and the studies in the following section round out the picture (Practical Ecommerce).
Google notes that AI responses may reflect what is said about products and services on blogs, videos and forums (TechWyse). Genuine mentions help. Fabricated mentions risk being treated as spam.
Two studies by Ahrefs help to distinguish between what works and what merely appears to work.
Ahrefs analysed 15,000 prompts across four AI assistants. On average, only 12 per cent of the links cited by ChatGPT, Gemini and Copilot appear in Google’s top 10 results for the same prompt. Perplexity is the AI assistant most closely aligned with SEO results. In AI Overviews, previous studies by the same company indicate that 76% of citations come from pages in the top 10 (Ahrefs).
The practical takeaway: a good ranking on Google helps in AI Overviews and Perplexity. ChatGPT and Gemini draw on other sources, which is why presence outside the website itself carries more weight.
In May 2026, Ahrefs published a controlled study on JSON-LD. It tracked 1,885 pages that added schema between August 2025 and March 2026 and compared them with 4,000 control pages. The analysis covered Google AI Overviews, AI Mode and ChatGPT, over a 30-day period before and after the change (Ahrefs).
The result: no significant increase in citations on any platform. Prior to the study, an analysis of 6 million URLs had shown that pages cited by AI have around three times as much JSON-LD. The controlled experiment shows that this link is a correlation and that schema alone has little effect on the outcome (Search Engine Journal).
There are two limitations to note:
The recommendation: maintain the correct schema for the sake of rich results and technical clarity. Invest the extra time in content and external presence.
AI assistants form an opinion about a brand based on various sources. Google confirms that responses may reflect what is said on blogs, videos and forums. The Ahrefs study, which found only a 12 per cent correlation between mentions and top 10 rankings, points in the same direction: responses draw on a wide range of pages.
Genuine mentions build reputation. Google warns that fabricated mentions may be treated as spam. The safest approach is to give third parties good reasons to talk about the brand: original data, real-life case studies and useful opinions.
The academic study that coined the term GEO, presented at the KDD 2024 conference, showed that certain changes to content increase visibility in generative responses by up to 40 per cent (Princeton). The changes tested involved adding statistics, source citations and quotes from experts (arXiv).
The result comes from a controlled test environment and varies depending on the topic. It serves as a guideline and confirms what Google’s guidelines call for: useful, original and well-organised content.
The studies consulted for this guide use international data, and Portuguese-language prompts are awaiting the first major study. A company that publishes verified data on the Portuguese market is filling a gap in the market.
A page can only be cited if the relevant crawler can read it. Each AI company uses separate crawlers for three functions: search, model training and direct user requests (The Rank Masters).
| Platform | Search crawler | Function |
|---|---|---|
| Googlebot | Index that powers AI Overviews and AI Mode | |
| OpenAI | OAI-SearchBot | ChatGPT search index |
| Anthropic | Claude-SearchBot | Indexing for Claude search |
| Perplexity | PerplexityBot | Indexing for Perplexity search |
Sources: Cloudflare and GEOToolbox.
OpenAI states that websites which block the OAI-SearchBot are excluded from ChatGPT’s search results. A change to the robots.txt file takes around 24 hours to take effect for this crawler.
Crawler rules change frequently. It is always advisable to check the official websites of each company before making any changes to the robots.txt file.
Assistant responses vary from day to day. A fixed method, repeated every month, turns this variation into a trend. The method below is a working recommendation.
Decide how many prompts you wish to track. For example:
ChatGPT, Gemini, Claude, Perplexity and Google (AI Overviews and AI Mode). Repeat each prompt three times and record the results in a spreadsheet.
| Metric | What it shows | How to calculate |
|---|---|---|
| Mention rate | Frequency with which the brand appears in the response | Responses containing the brand ÷ total number of responses |
| Citation rate | Frequency with which the domain appears as a source | Responses containing a link to the domain ÷ total |
| Share of voice | Brand weight relative to the competition | Brand mentions ÷ mentions of all brands |
| Pages cited | What content works | List of the most cited URLs |
| Referral traffic | Visits from assistants | Sessions by source in web analytics |
HubSpot’s AEO module automates part of this monitoring: brand visibility, mentions, citations and share of voice compared to competitors, with details by assistant.
Figures for a single month on their own are not very informative. Comparing months shows the trend. An increase in mentions in comparison prompts is the first sign that your external presence efforts are paying off.
How many times has your brand been mentioned today?
We measure your visibility across the leading AI assistants and link it to your CRM, right through to qualified sales meetings.
For a B2B company, visibility is worth what it generates: leads, conversations and sales meetings. Linking these four stages allows you to understand the real return on investment.
A monthly report can bring together four metrics to track the evolution of AI visibility and its impact on the business: mention rate, referral sessions from AI assistants, leads originating from these channels, and qualified meetings.
Analysing these indicators together helps to understand whether increased visibility translates into traffic, commercial opportunities and results. Progress should be analysed over time and in conjunction with other factors that influence customer acquisition.
A 90-day plan organises the work into three phases: diagnosis, implementation and evaluation. Over this period, it is possible to test hypotheses, introduce improvements and monitor the first signs of progress.
The duration of each task and the number of pages, prompts or actions should be tailored to the size of the website, the resources available and the business objectives.
Please note: 90 days may be sufficient to implement changes and identify initial signs of progress, but they do not guarantee SEO or GEO results. Not all actions have an immediate effect, so monitoring should continue over the following months.
Visibility in AI-powered search is built on three ingredients: technical access, original content and reputation on external sources. Google confirms that the fundamentals of SEO play their part. Studies by Ahrefs show that other assistants benefit from a broader presence. Monthly measurement links everything to the business objective.
Those who start now have a clear advantage. Data on the Portuguese market is still scarce, and every company that publishes verified information gains ground.
Turn this guide into a plan for your business
We’ll assess your AI visibility, set priorities for the first 90 days and link the results to qualified sales meetings.
How this article was produced
We wrote this article with the support of artificial intelligence and it underwent human review: one expert reviewed the text (copy) and another validated the content from a technical perspective. The sources are cited throughout the article and in the list above. Editorial responsibility lies with SmartLinks.