Google's AI content advice for publishers chasing visibility

What happened

Per Search Engine Journal, Google's Director of Software Engineering, Paul Haahr, told an audience at a recent industry event that publishers should use AI "in the best possible way" to add value, and that this is "something we can advise." The remark came during a discussion of how AI-generated content interacts with Google's ranking systems and AI Overviews.

Haahr's framing is a notable shift in tone. Google spent two years telling publishers it does not penalise AI content per se, only low-quality content. Now a senior engineer is going further: use AI to make the work better, and Google will treat that as a positive signal. The implicit message is that AI assistance, applied well, is becoming the expected baseline, not a risk to manage.

The advice sits inside a broader question publishers keep asking: what does Google actually want when its own answer engine is ingesting their pages to generate AI Overviews and feed Gemini?

Why it matters for your brand

For B2B brands, this is permission to stop treating AI in the editorial workflow as a liability and start treating it as a quality lever. That matters more than it sounds. Most large enterprises, multilaterals, and policy institutions we work with still operate under internal guidelines that effectively discourage AI in content production, written when the reputational risk of a hallucinated stat outweighed the productivity gain. Those guidelines are now out of step with how the platform doing the ranking thinks about the work.

The practical implication for visibility in AI search: Google's AI Overviews and Gemini cite pages that demonstrate expertise, structure, and clarity. AI used well, for outlining, fact-checking against primary sources, tightening prose, generating structured summaries, improves all three. AI used badly, for generating undifferentiated bulk, hurts all three. The gap between brands that figure out the first pattern and brands stuck on the second will widen quickly.

For financial services, the calculation is specific. A global bank's research function publishes hundreds of market notes a year. Most never rank, and almost none get cited in AI answers, because they read like internal memos pushed onto a CMS. AI-assisted editing that adds clear definitions, structured takeaways, and explicit comparisons to prior periods would not dilute the analyst's voice; it would make the page legible to the retrieval layer. The same logic applies to insurer thought leadership and asset manager outlooks.

For multilaterals and the UN system, the constraint is different but the opportunity is larger. UNDRR, World Bank groups, OECD: these organisations sit on enormous volumes of authoritative research that LLMs already trust. The bottleneck is surface area and structure, not credibility. AI-assisted production of explainers, FAQ layers, and translated executive summaries is the fastest way to multiply citation surface without diluting the underlying work. Google's signal here removes the last institutional excuse to not do it.

For major industrial groups and philanthropic institutions, the play is reformatting. Annual reports, sustainability disclosures, and policy white papers are written for regulators and boards, not retrieval. AI can convert them into the question-shaped content that AI Overviews actually pull from, without rewriting the underlying claims. A Holcim sustainability metric or a Gates Foundation programme outcome buried on page 47 of a PDF is invisible to Gemini. The same fact, surfaced as a structured answer to "how much has X reduced Y since 2020," is citable.

The risk that remains: Google did not say AI content is safe. It said AI used to add value is advisable. Brands that read this as a green light to scale generic output will lose visibility, not gain it. The differentiation will come from how editorial judgment is applied to AI drafts, not from whether AI is in the workflow at all.

The signal in context

This is the third major signal in 2024 to 2025 that the platforms shaping AI search have shifted from "AI content is suspicious" to "AI content is fine, quality is what we measure." OpenAI's content partnerships with publishers, Anthropic's expanding citation behaviour in Claude, and Google's repeated clarifications on its helpful content system all point the same direction. The competitive question for brands is no longer whether to use AI in production. It is whether the resulting work is better, more structured, and more citable than what competitors are producing with the same tools.

The brands that will win visibility in LLM answers over the next 18 months are the ones rebuilding their content operations around two questions: does this page answer a real question better than anything else on the open web, and is it structured so a retrieval system can extract the answer cleanly. AI is the production lever that makes both feasible at scale. Google's engineering leadership now saying so out loud is the cover that internal teams have been waiting for.