The European Commission has formally designated ChatGPT a Very Large Online Search Engine under the Digital Services Act. For marketers, this is a reason to take AI discovery seriously, not a reason to throw away an existing SEO strategy.
The useful response is to ask how customers research your category, whether your business is represented accurately, and how you would know if visibility produced a commercial result.
Reviewed 8 September 2026. This article focuses on marketing implications, not a legal compliance assessment.
What the EU actually announced
The Commission's 31 August 2026 announcement designates ChatGPT as a VLOSE. The announcement says the designated services reported at least 45 million average monthly EU users. Additional DSA obligations apply to those services following notification, including systemic-risk assessment and mitigation.
This is a regulatory designation. It does not publish a ranking formula, certify an SEO service, or promise that an individual website will be recommended. Those are different questions from the legal category assigned to the service.
Our marketing interpretation is narrower: include AI search in channel planning where it is relevant to your buyers, and require evidence before increasing investment.
Do not confuse channel recognition with a growth guarantee
A new channel can matter without deserving the same budget as your best-performing channel. We suggest beginning with a small, repeatable investigation rather than a large speculative content programme.
For a business selling training, the questions might concern course suitability, accreditation details or local availability. For a software product, they might concern integrations, alternatives, pricing structure or fit for a particular workflow.
Use actual buyer language from enquiries, sales calls and support conversations. A prompt written only to make your own product win tells you little about genuine discovery.
Build a buyer-question map
Our recommended starting point is a compact map covering three stages: understanding the problem, comparing options and checking whether a particular business is suitable.
For each question, record the audience, market, language and desired next step. Keep branded and unbranded questions separate. Asking an assistant about your named company tests something different from asking it to recommend companies without prompting your name.
Choose a repeatable set that your team can maintain. Ten well-chosen questions with saved evidence can be more useful for decision-making than hundreds of poorly labelled screenshots. That is a workflow recommendation, not a statistically representative market sample.
Include the less flattering questions too. "Who is this product not suitable for?" may reveal a more valuable content gap than another broad "best software" prompt.
Check discovery access without assuming inclusion
OpenAI's publisher guidance says OAI-SearchBot access supports eligibility for search summaries and snippets. Allowing access does not guarantee a recommendation. Training permission is a separate issue.
Ask your technical team to check the intended access configuration and document any restrictions. Then examine the pages a prospective customer would need in order to make a good decision.
We recommend clear explanations of what you offer, who it serves, where it is available and what evidence supports the claims. Show meaningful limitations and conditions instead of filling a page with synonyms for "best". That improves the usefulness of the information whether the reader arrives from an assistant, a search engine or a direct recommendation.
For Google's AI features, its Search Central guidance says established SEO fundamentals remain relevant and no special AI schema is required. Avoid buying a supposed universal shortcut based only on terminology.
Separate mentions, citations and recommendations
Use explicit definitions in your own evaluation. A mention means the name appears. A citation means the answer links to a relevant source. A recommendation means the response positively presents the business as an option for the stated need.
These are proposed measurement definitions, not a claim that every platform calculates them the same way. Save the underlying response so that somebody else can check your classification.
For example, an answer might mention a product only to explain why it is unsuitable. Counting that as a successful recommendation would make the dashboard look better while making the analysis worse.
Our AI citation tracking guide explains how to keep repeated observations comparable, including what to do when a model or search experience changes.
Set a sensible first-month objective
For an initial project, we would prioritise a baseline, a short list of factual content gaps and one or two useful improvements. Agree how the team will recognise an encouraging result and what would count as inconclusive.
Do not promise a certain number of customers from a handful of test responses. Track identifiable visits and qualified outcomes separately, and ask customers how they discovered the business when appropriate. Our AI attribution guide sets out that distinction.
The UpSearch AI Mention Tracker provides a relevant starting point for ongoing question-based monitoring. Use the observations to guide investigation, not to claim complete knowledge of every answer every customer receives.
The strategic opportunity is to become easier to understand and evaluate. The regulatory announcement makes the channel harder to dismiss, but your own evidence should still determine what you do next.
FAQ
Does the designation reveal how ChatGPT ranks websites?
No. The Commission's announcement concerns DSA designation and obligations for the service. It is not documentation of an SEO ranking system or a guarantee of website inclusion.
Should a small business replace Google SEO with ChatGPT SEO?
We recommend testing AI discovery alongside existing acquisition work. Use buyer relevance, observed results and commercial outcomes to decide whether more investment is justified.
What is the most useful first step?
Write down real customer questions and establish a repeatable baseline. That provides a clearer starting point than generating large amounts of content without knowing which decision it is meant to support.