Citable content for AI search: original, expert, verifiable

The Reuters Institute's 2026 survey found that 42% of people who use a chatbot for news ask follow-up questions, and the same share looks for more depth than a summary provides. For a business, the consequence is that citable content for AI search, the text a model can use as a source, is built from what the company knows and no one else can write. The generated answer opens a further question, and closing it takes expertise.

People using AI answers want depth

The Reuters Institute tracks chatbot use for news across 45 markets. The share of people using them weekly rose from 7% to 10% in a year, reaching 17% among 18-24s against 5% in the oldest group, and only 1% name them as their main source of news. The picture is of a complementary channel: the assistant sits alongside existing sources and is used to understand.

Trust explains the growth. Trust in news from chatbots stays at 20% overall but rises to 44% among users. The motivations point the same way: 42% want more depth or explanation, 39% want speed, 33% ask for a judgement on a source's reliability. People who turn to this route are looking for a competent counterpart on a question the summary left open.

Google's AI features trigger on the same pattern. Pew Research Center, looking at almost 69,000 searches collected between March and April 2025, found that a summary appears about once in five, and far more often when the search is a question (60% of cases) or runs to ten words or more (53%). These are the developed queries, where the user already has context, and they resemble the questions a potential customer asks. Within the four levers of digital visibility, this channel joins content, advertising and conversion; the difference is that presence here is won as a source.

Content that cannot be replicated

Google published a guide to its AI features in May 2026. The central passage concerns the quality of the text: content readers find useful and original weighs, over the long run, more than any other optimisation. Google calls it "non-commodity", meaning text that carries something exclusive: a point of view of its own, first-hand experience and expertise that goes beyond common knowledge. At the other end sits generic content, the kind that summarises what is already available elsewhere and that a model would produce on its own: it is the part of the web that can be replicated without losing anything.

The guide also lists what is not needed. Google says it ignores `llms.txt` files and AI-oriented markup, does not ask for text to be broken into chunks, does not require content to be rewritten for generative features and does not reward mentions built for show. Structured data remains useful for rich results, but it is not a requirement to appear in AI answers. The usual conditions hold: an indexable page, content accessible to crawlers, a readable structure.

The room for shortcuts narrows. Google warns that no third-party tool has access to its internal systems, and that foundational SEO remains the basis even when a model writes the answer. Optimising for AI features, for Google, is optimising for search.

From first-hand data to a citable source

AI features retrieve pages from the index and use them to ground the answer, through retrieval-augmented generation, and issue related searches around the same question. The model selects and connects what it finds. For a business, the work is to make its material selectable, and the selectable part is the one that brings an original contribution.

Content a model cannot reconstruct comes from experience: the data of a project, the real timings of a process, a case with verifiable numbers, a method described in full. Pew Research Center finds that the sources most often cited in AI summaries are Wikipedia, YouTube and Reddit, and that institutional sites are over-represented compared with traditional search. General knowledge is already covered; the space that remains is specialist, documented by those who lived it.

Coherent signals sit alongside the content. A recognisable author, a date, an identifiable company and information aligned across website, profiles and external mentions help the system connect the text to a real subject. The work on reputation and external citations acts on that level, with reviews, relationships and links; the content the model extracts is the part the company controls directly.

Measurement in AI features

Since June 2026 Search Console has shown a report dedicated to AI features, extended to all sites on 31 August. The report measures impressions: how often a link from the site was shown in AI Overviews and AI Mode, broken down by page, country, device and date. Clicks have no separate line: traffic from AI features stays inside the overall search figure and has to be read alongside site analytics.

The report's scope is worth knowing. Google does not include data from Search Labs experiments and warns that the most recent figures can be provisional. The numbers circulating about traffic loss belong to the news industry: according to Chartbeat data cited by the Reuters Institute, organic Google search traffic to more than 2,500 news sites fell 33% globally and 38% in the United States between November 2024 and November 2025. Publishers expect a further 43% reduction over three years. These are forecasts for one specific sector, not a measure that holds for every business.

The control levers are the usual ones. Crawlers are governed with `robots.txt`, content visibility with `nosnippet` and `max-snippet`, training in other Google systems with `Google-Extended`. Since June 2026 Google has been testing a control to opt out of AI features, initially in the United Kingdom: sites that opt out receive no traffic or impressions from those features, and the choice does not affect ranking in traditional search.

A method for a small business

For a small business the starting point is an inventory of what it knows. The questions customers ask before buying, the data the company collects as it works, the cases it can document without breaching a client's confidentiality: this is material a model does not find elsewhere. From there it chooses what to make public, following market demand.

The second step is verifiability. Content an AI system can cite carries numbers with their source, dates, an identifiable author and a repeatable method. Coherence between what the website says, what the company states elsewhere and what customers report makes the subject recognisable. Digital advertising in the age of AI covers immediate demand and remains a separate channel; organic presence in answers is built over time.

The technical choice comes later. A generic subscription to a tool is a starting point; a setup that connects data, content and measurement takes design and maintenance, and produces material that accumulates. The difference shows over time, not in a single publication.

Measurement closes the loop. Impressions in AI features show how present the site is; contacts and conversions show whether that presence produces value. Rising impressions with flat contacts point to a content or journey problem, while rising contacts with stable impressions point to work on conversion. Each figure has to be read with the other, and the answer arrives over months.

The advantage of a small business in generative search lies in what it knows and has yet to document. What AI produces on its own is the repeatable part of the web; the rest is experience, and it stays with whoever holds it.