AI search engines don’t hand you ten blue links. They read the web for the user, stitch an answer together, and let them keep asking follow-ups without ever starting a new search. That single behavioural shift is quietly rewriting how your customers discover brands.
The Semrush Blog ran a hands-on test of nine of these tools, using a consistent buying scenario
try: accounting software for a small US marketing agency, with a set budget and feature list. The reviewer asked follow-ups, opened the citations, and cross-checked details against the actual vendor websites. That last step is where things got interesting for marketers.
The nine tools on the list
- Google AI Mode — strongest for product and local searches, with prices, reviews and nearby inventory. Free basic access, paid from $9.99/month.
- ChatGPT — best for research chains with lots of follow-ups; deep research lets you edit the plan and prioritise sources. Paid from $8/month.
- Perplexity — best for academic and premium sources like Wiley, Statista, PitchBook Essentials and CB Insights. Paid from $20/month.
- Microsoft Copilot — shines inside Edge, where it can use your open tabs, PDFs and browsing context, plus price tracking.
- Brave Leo — private, no account needed, and you can bring your own model.
- Duck.ai — private, no account, multiple model providers, and a handy “2nd opinion” switch.
- Claude — granular Effort controls (Low through Max) and interactive clickable follow-ups.
- Consensus — 220 million-plus peer-reviewed papers, updated weekly, with a Consensus Meter showing whether findings lean Yes, No, Possibly or Mixed.
- DeepSeek — fully free with DeepThink reasoning and a 1 million-token context window.
The accuracy gap is your opportunity
The most useful finding for brand teams isn’t which tool won. It’s how often the pricing was wrong.
Google AI Mode, per the test, in some cases showed a product’s cheapest price next to features that only came with pricier plans. Brave Leo served a couple of outdated figures. DeepSeek repeatedly returned stale pricing and leaned on third-party sites rather than vendor pricing pages. Perplexity, Copilot and Claude held up better on the details that were checked.
If your pricing lives inside a JavaScript-heavy component, a PDF, or a “contact sales” wall, an AI engine will happily fill the gap with whatever a review site said 18 months ago. That misinformation then becomes the answer your prospect sees before they ever hit your site.
Who actually gets cited
Across the tests, the reviewer noted that Claude leaned repeatedly on the same handful of software marketplaces and comparison sites (G2 showed up again and again). Brave Leo pulled mostly from third-party review sites and marketplaces, with official vendor sites largely absent. ChatGPT and Perplexity drew on the widest mix, including brands’ own websites.
Translation: for a big chunk of AI search, your third-party footprint is your brand. Your G2 profile, your listing on comparison roundups, your presence in “best tools for X” articles — those are the pages doing the talking.
A four-step AI-visibility checklist
Here’s how to act on this now:
1. Audit your own machine-readable facts. Pricing, plan inclusions, integrations and availability should be plain HTML text on a crawlable page, with structured data. No images, no gated PDFs.
2. Own your third-party listings. Claim and refresh every review-site and marketplace profile. If a marketplace has your old price, an AI engine may quote it.
3. Test the answers yourself. Run your category’s buying prompts across ChatGPT, Perplexity, Google AI Mode and Copilot. Note whether you appear, which sources are cited, and what’s factually wrong. Repeat monthly.
4. Write for the follow-up. These are conversations, not queries. Buyers narrow by budget, team size and features. Publish pages that answer those specific slices — “pricing for teams under 10,” “cheapest plan with X integration” — because that’s the shape of the second and third question.
Why it matters
Zero-click was already eating organic traffic. AI search adds a second problem: the answer can be confidently wrong about you, and there’s no obvious place to appeal. The brands that win here are the ones making the truth about their product the easiest thing on the internet to find and quote.
Source: Semrush Blog
