AI Search Prompts Push Claude and Perplexity Against Google

A prompt list that says something bigger

A recent Tom's Guide column called "Forget Google searches" gives readers seven prompts meant to replace the usual keyword search. The writer says that Google's AI Overviews and its old pages of blue links pushed her toward chatbots, which she found gave more personal results.11 The prompts are long and full of conditions. One is a $500 home-office upgrade request that takes into account the gear she already owns. Another ranks wireless headphones and drops any model that misses a single requirement. A third asks the chatbot to question the user before deciding whether a handheld gaming PC is worth buying.11

The column mostly tests these prompts in Gemini and ChatGPT. It mentions "other like-minded chatbots" only in passing.11 That gap matters, because the two companies whose products are built most directly around cited, research-style answers are Anthropic and Perplexity. The bigger question is not whether a smart prompt beats a search box. It is which assistants are actually winning the users who are making that switch, and the numbers here are less tidy than the column's tone suggests.

The prompts point toward research tools, not simple chat

The seven prompts have one thing in common. Each asks the assistant to collect current information, compare choices against stated criteria, and explain its reasoning. Those are research jobs, not quick lookups. The writer even tells the model to "research current options" and "research current models, prices, brightness" before reaching a verdict.11 This is the kind of work Perplexity has built its business on, and it is also why Anthropic added live search to Claude.

An earlier Tom's Guide test makes the same point from Perplexity's side. In a seven-prompt matchup with Google's AI Overview, Perplexity wrote full answers where Google sometimes fell back to ordinary links or thin snippets. On a Tokyo itinerary request, Google offered mainly third-party travel sites, which the reviewer noted also helps Google's ad revenue.15 Industry analysis lists Perplexity's citation density as a clear strength, at five to fifteen numbered references per answer, placed at the top.30

The two pieces disagree in a telling way. The 2025 comparison treated Perplexity as the natural option for people moving away from Google.15 The 2026 prompt list doesn't mention Perplexity at all and gives the job to general-purpose assistants.11 That change in coverage lines up with what the traffic data shows.

Perplexity: making more money while losing share

The numbers on Perplexity point in different directions, and it matters which one you trust. Several trackers agree that its share of chatbot referral traffic has fallen sharply. Statcounter's data, as reported by several analysts, shows Perplexity falling from a 12.07% peak in April 2025 to 4.31% in August 2026, while Gemini rose to 10.9%.2226 Similarweb's measure is even lower: 0.9% of global generative-AI web traffic, down from 2.0% a year earlier.22 Several reports say Gemini passed Perplexity as the second-largest AI search referrer in January 2026.2530

User numbers vary much more. Estimates include 34 million monthly active users as of March 202627, about 45 million on the core answer engine with more than 100 million across all products22, and a mid-2026 figure of more than 100 million25. Reports also put query volume above a billion a month.22 The best reading is that Perplexity's user base is still growing in absolute terms, but the whole market is growing faster, so its share is shrinking. One analysis calls this "share dilution" in a market that grew 36%.30

The business side looks healthier. Perplexity's annualized revenue is reported at around $750 million, and it was valued at $22.6 billion in January 2026.27 Another analysis says the next valuation could reach $30 billion or more, calling Perplexity "a small platform with a big business."22 One review claimed Perplexity had taken 15% of the traditional search market in early 2026.23 That claim doesn't fit any of the measured data. Statcounter has Google at 91.1% of global search in August 2026, and most reports say Perplexity is still too small to show up in search-share rankings at all.2427 The 15% figure looks like marketing, not measurement.

Anthropic: few users, high value

Anthropic's story is close to the reverse. Claude has had live web search since a paid, US-only preview in March 2025, and it reached every plan worldwide, including the free tier, by late May 2025.4 Users turn on search from the tools menu in the chat box. Claude then decides by itself when a question needs current information and cites its sources inline.2 A separate Research mode for paid plans runs for minutes, pulls from the web and Google Workspace, and produces long reports with citations.23 As of February 2026, it can also connect to MCP servers that hold company data.3

That setup suits the Tom's Guide prompts well. A request like "compare 10 AI voice recorders and rank them" is the kind of multi-source job Research mode is designed for. Claude's ability to ask questions before answering also fits the column's seventh prompt, the "answer the question behind my question" approach.11

As a traffic source, Claude is still small. Statcounter data puts it at 2.6% of AI chatbot referrals in August 2026, behind Copilot.26 Who uses it tells a different story. One analysis finds Claude's share of B2B referral traffic is 18.5%, about seven times its share overall.30 Its conversion rate is reported at 16.8%, the highest of the major assistants.30 Growth estimates vary a lot, from 320% year over year25 to 770%30. Both start from a small base, and the exact number depends on the method. The direction is clear, though: Claude is gaining among business users with high-value research needs, which is roughly the kind of user the Tom's Guide prompts are written for.

There is one technical detail people who prefer Claude to Google may not know. Several independent signals suggest Claude's search runs on Brave, not Google or Bing. Anthropic lists Brave Search as a web-search subprocessor, and one study found 86.7% overlap between Claude's citations and Brave's top results.24 Google's own cloud documentation for running Claude models also names Brave Search as the third-party provider.5 So a Claude user who "forgets Google" really is using a different index.

On the developer side, Anthropic keeps building. The API web search tool costs $10 per 1,000 searches plus token costs8, and code execution became free when paired with web search in February 2026.1 October 2026 updates added domain restrictions for search in managed agents and new API flags showing which models support search.19

The catch: the same prompt doesn't always give the same answer

The prompt-list format suggests that the right wording gives reliable results. The research is more complicated. A paired experiment on 180 multi-turn conversations found that replaying a prompt without its earlier conversation changed 44.7% of answers. In 26.7% of cases the recommendation or conclusion itself changed.4 Other work found that AI assistants recommending brands mostly cite third-party company pages, with listicles alone making up 21% of citations.4 When a reader asks for the "best" headphones, the answer may come from the same affiliate-heavy pages that made traditional search frustrating in the first place.

The disagreement among trackers is also a reason for caution. Perplexity's usage share was 7.73% in April 2026 and 4.3% four months later on Statcounter's measure2826, while Similarweb puts it under 1%22. Any claim that one assistant has "won" AI search depends on which dataset you use.

What it adds up to

The shift away from Google that the column describes is real, but it is a minority habit, not a mass exodus. Google still sent 91.1% of search-engine referrals worldwide in August 2026. One analysis argues that the bigger threat to publishers is Google answering more queries itself, not rivals taking its users.26 Even among chatbots, ChatGPT sends about 79% of AI referrals.26

For Anthropic and Perplexity, the fact that the prompt list barely mentions them shows the strategic problem. Both companies built the cited, research-first experience that these prompts depend on. Yet general-purpose assistants, Gemini in particular with its built-in distribution, are picking up the casual users who are leaving keyword search. Perplexity is earning more while its share shrinks. Anthropic is staying small overall but doing well with business users. My view is that Anthropic's approach looks more durable, because the readers writing detailed, research-style prompts behave a lot like the business users Claude is already attracting. Perplexity's problem is that this research style is no longer unique to it.