
A local media outlet cannot compel Google, ChatGPT or another AI system to cite a particular article. A newsroom can, however, substantially improve its chances: make pages accessible to search, state facts clearly, identify sources, publish original local information and show who takes responsibility for the reporting.
This guidance draws on the National Union of Journalists of Ukraine’s Optimise or Disappear article and its webinar Can AI Be Made to Cite Media?, featuring journalist and Academy of Ukrainian Press media trainer Andrii Yurychko.
Can citation be guaranteed? No. Neither a metatag, a special file nor a brand mention guarantees inclusion in AI Overviews, AI Mode, ChatGPT or another generative response.
What improves the chances? Technical accessibility, a clear headline and opening, precise attribution, a stable page address, identifiable authorship and the newsroom’s own data.
Does GEO replace conventional SEO? No. Google’s official guidance says the basic SEO principles still apply to generative search. There are no separate technical requirements for appearing in AI Overviews or AI Mode.
A small survey of registered webinar participants found that 97% of newsrooms used AI tools regularly or occasionally, yet only 8% could easily find their own reporting in Google and track its use by AI systems. This was not a representative study of all Ukrainian media, but it points to an important gap between using technology and measuring visibility.
NUJU Project Activities Coordinator Iryna Khromenko highlighted the changing search model: instead of a list of websites, users increasingly receive a ready-made short answer. For media, this means competition for attention, website visits, advertising and subscriptions.
‘Repetitive generated texts increase the value of quality original journalism,’ Yurychko stressed.
AI can speed up transcription, initial structuring, identifying contradictions and drafting alternative headlines. A newsroom’s main advantages remain reporting on the ground, local memory, verified documents, its own data and responsibility for every published fact.
Articles should open without sign-in, work properly on smartphones and not be blocked by robots.txt or a noindex metatag. Newsrooms should check indexing in Google Search Console, their sitemap and security service settings.
To make pages eligible for ChatGPT summaries and snippets, OpenAI advises against blocking OAI-SearchBot. GPTBot concerns potential use of content for model training and is controlled separately.
The headline, opening and first paragraph of each section should quickly explain what happened, where, when, to whom and why it matters. ‘The city council’, ‘in autumn’ or ‘allocated 1.8 million’ is insufficient without the community’s name, year, currency and decision number.
An important fact should retain its attribution even if a search engine displays only one paragraph. Name the institution, document, official, data author or newsroom that obtained an exclusive response. This protects reporting from inaccurate summaries and supports proper citation.
A system can display a short answer without sending the user to the website. An article therefore needs additional value: the full document, a map, table, timeline, calculator, original photographs or video, methodology, or an explanation of consequences for that particular community.
Google separately emphasises useful, original, non-formulaic content created primarily for people. Its official guide to optimisation for generative search features explains this further.
hreflang relationships.Google does not require an llms.txt file, a separate ‘machine’ version of an article or special schema markup for AI Overviews. There is no need to split every text mechanically into dozens of short blocks or create pages for every possible query variation.
Nor are there guaranteed formulas such as ‘the first 100 words are all that matters’, ‘update a third of the text’ or ‘add a map and AI will bring traffic’. Structure, factual accuracy and useful multimedia matter when they help readers, not when they merely imitate optimisation.
Do not assess visibility through a single random query: results depend on wording, country, time, personalisation and search mode. Google Search Console remains the basic tool.
In June 2026, Google announced the gradual introduction of separate reporting on impressions in generative Search and Discover features. These reports show pages, countries, devices and changes in impressions. ChatGPT referral traffic can be tracked using utm_source=chatgpt.com.
NUJU organised the practical webinar Can AI Be Made to Cite Media? Its speaker was journalist, AUP media trainer and lecturer at the Educational and Scientific Institute of Journalism, Taras Shevchenko National University of Kyiv, Andrii Yurychko. The event formed part of NUJU’s priority area, Restoring Ukrainian Media, supported by Ukrainian Media Fund Nordic.
For AUP, this subject is part of its sustained work in professional journalism and responsible use of digital tools. Read more in Media Literacy 2.0: Using AI for Media Literacy and Education.
Source: NUJU Information Service.
Translation note: This is a literary translation into British English, produced and proofread using a language model. Read the Ukrainian original.