SEO for AI search engines: how to get ChatGPT and Perplexity to cite your brand

What is SEO for AI search engines?

SEO for AI search engines, also called AI search optimisation, is the set of techniques that increase the likelihood of an assistant mentioning a brand or citing its pages when it answers. It is also known as GEO, short for Generative Engine Optimization. It applies to ChatGPT, Perplexity, Gemini, Copilot and Google’s AI Overviews.

The difference from traditional SEO lies in the outcome being pursued. On Google, you compete for a position in a list of links. In an assistant, the answer already resolves the question, and visibility depends on appearing within it as a recommended brand or as a linked source.

For a business, the shift mainly affects evaluation queries: supplier comparisons, alternatives to a tool or recommendations for a specific use case. These are searches with purchase intent in which the user usually asks for a shortlist of options.

How does it differ from traditional SEO?

Both disciplines share the same technical foundation: crawlable pages, useful content and a brand with authority. What changes is the weight of each factor and the measure of success. Traditional SEO measures rankings and clicks; SEO for AI measures how often the brand is mentioned or cited and whether the assistant describes it accurately.

The two channels feed each other. Many assistants retrieve information from the Google or Bing indexes, so a poorly indexed page will not make it into their answers either. That is why every SEO for AI project begins with a review of technical and content SEO.

How do AI search engines choose their sources?

AI search engines combine two routes. The first is what the model learned during training, which is only updated with each new version. The second is a real-time search. The assistant rephrases the question into several queries, retrieves pages from an index and writes its answer from the most useful passages.

Influencing training takes time and depends on new versions of the model. Real-time search can reflect changes within a few weeks, provided the page is indexed, the crawler can access it and it contains a passage that answers the query.

Which crawlers to allow

Each assistant accesses the web with its own user agents, and several of them separate crawling for model training from crawling to answer searches. In the robots.txt file these are independent decisions. A company can exclude its content from training and still appear as a cited source.

What factors influence whether an AI cites a brand?

Assistants favour sources they perceive as reliable and easy to extract from. In practice, four factors carry weight. The first is mentions on third-party sites; next come citable content, a consistent brand entity and technical access to the pages.

Mentions on third-party sources

The more independent sources associate a brand with a topic, the more likely the assistant is to include it in an answer on that topic. Industry media, supplier comparisons, specialist forums, professional associations and review platforms all count.

In Spain, it pays to prioritise Spanish-language sources, because a query written in Spanish tends to retrieve pages in that language; the same logic applies to every market and language. This part of the work has more in common with digital PR and media relations than with buying links.

Citable content

The assistant cites passages, almost never whole pages. A section that opens with a clear answer, provides specific data and makes sense without reading the rest has a far better chance of being selected. Texts that take four paragraphs to get to the point are rarely cited, however good they are.

A consistent brand entity

To recommend a company, the assistant needs to know what it is, what it sells, to whom and in which market. If the website says one thing, LinkedIn another and a directory still holds the description from five years ago, the answer will mix up the details or choose a better-defined competitor. Structured data for your organisation and services helps to pin that information down.

Technical access and freshness

A page the crawler cannot read does not exist for the assistant. The date matters too. For queries where timeliness is important, such as prices, regulations or tool comparisons, assistants tend to prefer updated content with a visible date.

A brand that nobody mentions outside its own website is unlikely to appear in an assistant’s recommendations, however good its services page is.

Juan Berges, CEO of The Baller Company

How to structure content so it gets cited

The structure that works best in assistants is the same one a reader in a hurry appreciates. It is the same approach we follow when planning a content and brand strategy.

1. Answer in the first two sentences of each section, with a 40 to 60-word paragraph that resolves the question without preamble.
2. Phrase headings as the questions customers ask on sales calls and in emails.
3. Write self-contained blocks, so that each section makes sense on its own, with no references to earlier sections.
4. Provide your own data with units and dates: timeframes, price ranges, requirements or thresholds the company knows from experience.
5. Set out comparisons in tables, with explicit criteria and one row per option.
6. Name entities in full the first time they appear, whether the company, the product, the city or the regulation.

Example: a service page before and after

Take as an example a data protection consultancy based in Valencia. Its service page opens with three generic paragraphs about privacy. Its clients’ main question, how much it costs to bring a small business into line with the GDPR, does not appear until the end.

1. Open with a definition of the service that states what it includes, what type of company it is for and where it is provided.
2. Add a paragraph with the indicative price range and the usual timeframe for achieving compliance.
3. Include a table comparing the initial compliance project with annual maintenance.
4. Turn the questions that arrive by email into headings, each with its own answer.

With those changes, the page offers passages an assistant can use as they are for a query such as “GDPR compliance cost for small businesses in Valencia”. The previous version had none.

Common mistakes we find in audits

When we review how brands appear in assistants, the most frequent failings have little to do with the writing. They are almost always problems of access, of consistency between sources or of how measurement is approached.

• Crawlers blocked by the CDN or firewall, often after bot protection has been switched on without reviewing the exceptions.
• Main content loaded with JavaScript in the browser. Some AI crawlers do not execute scripts and receive an almost empty page, something that web development with server-side rendering avoids from the outset.
• Prices, terms or requirements published only in PDFs, images or videos without a transcript.
• Outdated external profiles, with an old trading name, services no longer offered or a previous head office.
• Service pages that do not say who they are for or in which market they are offered.
• Conclusions drawn from a single query, run on a single day in a single assistant.

How do you measure visibility in AI search engines?

Visibility in AI search engines is measured with a fixed panel of queries that is run periodically in each assistant. None of them provides a rankings report, so tracking combines that panel with the referral traffic recorded in web analytics.

How to build the query panel

A useful panel has between 20 and 30 queries spread across different intents. It should include informational questions about the sector, supplier comparisons, alternatives to specific competitors and direct questions about the brand itself. They are written the way a customer would write them, with context such as company size, country or budget.

Answers vary between sessions. That is why we run each query three times in a clean session, with no history and memory switched off. We note whether the brand appears, whether it links to one of its own pages, in what position and which competitors are cited. Repeating the measurement every month is enough to see trends.

Referral traffic and consent

In Google Analytics 4, it is worth creating a custom channel group with the assistants’ domains, because by default they are mixed in with other referrals. Some visits arrive without a referrer, particularly from the desktop and mobile apps, so the real figure is higher than the recorded one.

In Europe, the effect of cookie consent comes on top of that. With Consent Mode, sessions from users who decline are modelled or lost, so the trend is more reliable than the absolute value. To link that traffic to leads and sales, analytics and CRO reports should treat AI as a channel in its own right.

Where to start

We recommend measuring before changing anything, so that you have a baseline to compare against. After that, this sequence covers the first 90 days well:

• Define the panel of 20 to 30 queries and measure the brand alongside three competitors.
• Check crawler access in robots.txt, the CDN and the firewall.
• Restructure the ten pages with the most commercial potential into a citable format.
• Correct external profiles and organisation markup.
• Design a plan for mentions in the media and industry sources.

If competitors appear in the first measurement and the brand does not, the most urgent work usually lies outside the website.

Frequently asked questions