The investor stopped searching: what changes when the answer arrives synthesized

Search did not disappear. It changed shape. Instead of ten links to work through, the user gets one synthesized answer, supported by a handful of sources the model picked itself. A brand that is not in that handful does not lose a position; it leaves the conversation. This article gathers what can already be measured about that shift, with a declared source behind every number.

What changed in how people research financial information?

About half of US adults now use AI chatbots, up from a third in 2024, and roughly four in ten use them to search for information (Pew Research Center, fielded February 2026, 5,119 respondents). The habit is no longer a niche.

The Pew Research Center figure is the firmest evidence on adoption available today, because it comes from a probability panel with a declared sample: 44% of US adults report using ChatGPT, up from 34% a year earlier, and about a quarter say they use chatbots daily. This is not a habit confined to people who work in technology.

The two largest vendors disclosed numbers of the same order. OpenAI reported more than 1 billion weekly ChatGPT users in September 2026, and Google announced in August 2026 that the Gemini app had passed 1 billion monthly users. Both are company disclosures with no published methodology. Useful for scale, not as independent evidence.

The habit has reached the information diet. The Reuters Institute Digital News Report 2026, with roughly 2,000 respondents per market across 48 markets, measured 10% weekly use of AI chatbots for news globally, up from 7% in 2025, and 16% among people under 35.

One honest caveat, and it matters: as of September 2026 we could not find public research with a declared sample and methodology measuring AI use specifically among investors or financial-market professionals. The numbers above are general. We would rather state the gap than fill it with a secondary-blog estimate. It is the same rule we apply to client work.

The numbers on this page, with sources

Primary sources consulted on 9 September 2026. Where a figure is a company disclosure, the row says so. No number in this table was estimated by ReBo.

Why does a synthesized answer change the game for people who produce knowledge?

Because a list of ten links became one answer with a handful of sources. With an AI summary on the page, 8% of visits led to a click on a traditional link, against 15% without one (Pew Research Center, 68,879 real searches by 900 adults, July 2025). The click is no longer the scoreboard.

The July 2025 Pew study is the strongest evidence here because it measures real behavior rather than a Search Console sample: 68,879 searches by 900 US adults who shared their own browsing. When an AI summary was present, clicks on traditional links fell by roughly half. Clicks on links inside the summary accounted for 1% of visits.

Ahrefs measured the same effect from the publisher side. Comparing 300,000 keywords in Search Console between December 2023 and December 2025, the presence of an AI Overview correlated with a 58.0% drop in position-1 CTR. The loss concentrates at the top: at position 10 the drop was 19.4%. Whoever had most to lose was exactly whoever ranked first.

The practical consequence is that the scoreboard moved. Measuring success by organic sessions describes less and less of what is happening, because part of the audience received the company's information without ever opening the site. What matters now is whether the brand is read, understood, cited and trusted: the four verbs that organize all AI Visibility work.

If an AI cites few sources, how does it choose them?

Few indeed: an average of 7.7 domains per answer in AI Overviews, 7.3 in Perplexity, 5.0 in ChatGPT and 2.5 in Copilot (Profound, 100,000 prompts, July 2025). And the list barely matches Google's top 10: measured overlap ran between 20% and 26%.

The Semrush study is the only one we found measuring overlap between AI citations and organic results directly: across 200,000 keywords, coincidence with the top 10 ran between 20% and 26%, and more than 50% of AI Overviews on desktop and 60% on mobile did not link the first organic result. Note the date: collection is from September 2024, before AI Mode. The direction holds regardless: ranking well on Google is no passport to being cited.

The engines do not agree with each other either. In the Profound study of 100,000 prompts, ChatGPT and Perplexity shared 11.0% of cited sources; across the pairs tested, overlap ranged from 6% to 16.4%. In August 2026 the same firm measured that Claude and Claude Code, two products from the same company, mention only one in five of the same brands, across 24,135 answers.

That has a direct operational implication: measuring one engine measures one market, and optimizing for one is betting on a draw. What they share as a criterion is how easily a short, self-contained, verifiable passage can be retrieved, rather than where it ranks. The mechanism is detailed in how ChatGPT chooses the sources it cites, and the most common technical obstacle in why AI cannot read PDFs.

What does a company lose when it is not one of the cited sources?

Three things at once: the chance to be considered, control over how it is described, and any evidence that a problem exists. Absence does not produce silence: the AI answers anyway, using whoever was there.

The first loss is invisible by construction. Nobody complains, no alert fires, the traffic report does not change color. The gap only surfaces when someone runs the questions and logs who was cited. That is exactly why the baseline comes before any content production.

The second is control of the description. With no readable, dated source of its own, the model assembles an answer from whatever it found: an old registry entry, a third-party article, the description of a competitor with a similar name. Absence does not produce silence, it produces imprecision, and imprecision travels with the same authority as the correct answer.

The third compounds over time. Content cited today tends to be retrieved again tomorrow, because it joins the mesh of signals retrieval systems use to judge authority. The distance between those inside and those outside is not stable: it widens with every cycle.

None of this means a guaranteed position in AI answers exists. It does not, and we do not sell it. What exists is method and measurement: a set of target prompts, a baseline, a technical foundation that lets a crawler read the page, and a monthly cycle comparing citation rate against the starting point. That is what the maturity framework classifies across four levels, from Invisible to AI to AI Reference.