Perplexity AI in 2026: Ready Your Website for AI Agents

Perplexity AI spent the first half of August 2026 at the centre of two disputes that directly affect anyone with a website. It overturned in the US courts the injunction Amazon had obtained against its shopping agent, and it blocked ads that Time magazine was selling specifically to AI bots. This isn't Silicon Valley gossip: it's two signals of how traffic will arrive — or fail to arrive — at your site.

TL;DR

On 4 August 2026, the Ninth Circuit Court of Appeals overturned the injunction that prevented the Comet browser's shopping agent from accessing Amazon.

On 11 August, Perplexity AI blocked the markdown ads that Time was selling to AI agents, calling the practice deceptive.

In the UK, Perplexity AI is the second most used AI search tool, with about 8.42% of the market, behind ChatGPT.

Only 12% of AI citations match the top 10 on Google (Ahrefs). Ranking well doesn't guarantee being cited.

The practical work is technical and cheap: serve content in HTML, provide direct answers, use structured data, and measure referral traffic.

What Perplexity AI did in recent days

Two facts sum up almost everything that matters.

4 August 2026 — Amazon loses the injunction. The Ninth Circuit Court of Appeals in the United States overturned the decision that barred the shopping agent built into the Comet browser from operating within Amazon. The court's reasoning is the key point: it is the Comet user, not Perplexity AI as a company, who accesses Amazon's computers. Without unauthorised access by the company itself, the CFAA (the US computer fraud and abuse law) claim does not hold. Amazon had sent a cease-and-desist letter in November 2025 and sued in March 2026; the main case continues in the federal court in San Francisco. Source: Engadget.

11 August 2026 — ads made for bots, blocked. Time began selling advertising inserted into the markdown version of its pages, i.e. the plain-text version that agents read. The first clients were Ally Bank and the Project Management Institute. The response came quickly: Perplexity AI removed these ads from influencing its index. Jesse Dwyer, the company's head of communications, warned that outlets using “deceptive advertising like markdown ads” will lose reputation and trust within the system. Source: Digiday.

The backdrop of recent months helps read both episodes. In January 2026 the company signed a three-year commitment with Microsoft for $750 million in GPU capacity on Azure. In February it launched Computer, a general-purpose agent that carries out multi-step tasks on its own, and moved to a subscription-only model, ending advertising within the product. The company's valuation reached around $21.2 billion in the Series E-6 round, in early 2026.

Markdown ads: the fact that matters most for your website

This episode may seem like a media detail, but it's the best free lesson in AI optimisation of the year.

When an agent visits a page, it doesn't see your carousel, your popup, or your hero animation. It receives a text version — usually markdown or simplified HTML — and it is on that text that the answer is built. Time understood this and sold space there. Perplexity AI understood it too and cut it, because paid content presented as neutral fact contaminates the answer.

The lesson for an ordinary company is direct: the text version of your page is your page. It is from this that Perplexity AI builds the answer your customer will read. If the information that decides the purchase (price, delivery time, service area, condition, specification) only exists inside a component that depends on JavaScript, inside an image, or behind a click on a tab, it doesn't exist for the agent. And the way to fix this isn't to trick the bot — it's to serve the fact cleanly and verifiably. Those who try to manipulate get caught, as Time did.

Amazon's loss opens the door for agents buying on your site

The 4 August decision doesn't oblige anyone to do anything, but it removes the biggest legal scarecrow against agentic browsing in the United States. For Perplexity AI the effect was immediate: the Comet shopping agent resumed operating without judicial restriction. In practice, this accelerates a behaviour that was already growing: the user delegates the task and the agent enters the site, compares, fills in forms, and completes the purchase.

For those running e-commerce, catalogues, online quotes, or lead generation in the UK, three consequences deserve a place on the next technology meeting agenda:

Fragile flows become lost sales. Checkout that depends on hover, fields without label, aggressive captcha on quote forms, and steps that only work with human interaction now filter out not just bad bots, but real customers operating through an agent.
You need an explicit policy. Blocking everything by reflex is as bad as allowing everything without criteria. Decide what is public (product sheets, prices, policies) and what is restricted (logged-in areas, internal endpoints), and write this in your robots.txt.
Your logs become a source of intelligence. An agent's user-agent is identifiable. Knowing how many agent visits arrive, on which URLs, and with what error rate is the new metric that almost no one tracks yet.

It's the same movement we discussed in AI Agents: What Gemini Spark Means for Businesses — only now with a legal precedent reinforcing the agent's side.

How much does Perplexity AI matter in the UK?

Big press numbers often hide the number that matters. Let's look at the UK: among AI search tools used in the country, Perplexity AI holds second place.

Tool	Share of AI search in the UK
ChatGPT	Clear leader
Perplexity	8.42%
Gemini	5.87%
Copilot	3.77%
Claude	0.62%

Source: Full.services survey on the state of AI search in the UK in 2026.

Eight per cent may seem small until you look at the quality of the visitor. Someone arriving via an AI answer has already read a comparative summary and clicked because they want to confirm or close. It's bottom-of-funnel traffic disguised as discovery traffic. The Panorama PRO 2026 from Leadster measured a median conversion rate of 7.80% for the AI search channel — well above what most sites get from traditional organic traffic.

On the product side, the Comet browser is no longer exclusive to expensive subscribers and is now free on Windows, macOS, Android, and iOS, with a Chromium base and support for Chrome extensions. The 27 July 2026 update brought corporate roles and permissions, dedicated API credentials, memory (Brain) in 15 languages, Claude Opus 5 inside Search and Computer, plus verification features like Check Sources and the Source Context Panel. In other words: less toy, more work tool.

Citation is not ranking: what changes for your SEO

This is the most expensive mistake I see in marketing meetings. The team shows the first position on Google and concludes that it's sorted for AI too. It isn't.

Data from Ahrefs shows only 12% overlap between AI citations and the top 10 on Google. They are two different games with different signals. Being cited by Perplexity AI depends less on backlinks and more on your text containing the exact, dated, and easy-to-extract fact.

Dimension	Traditional SEO	AI optimisation (AEO)
Unit of competition	Position on SERP	Cited snippet in the answer
Typical query	3 to 5 words	~23 words, conversational
Dominant signal	Backlinks and domain authority	Factual clarity and consistency between sources
Winning format	Full page	Objective answer block
Success metric	Clicks and impressions	Citations, brand mentions, referral

There's also the already documented side effect: a TollBit report found up to 96% drops in referral traffic from AI search to news and blog sites, analysing 160 domains, as reported by Exame. Gartner projects a 25% drop in traditional search volume due to chatbots. In other words: the click becomes rarer and more valuable, and being cited gains brand value even without a click.

How to prepare your website for Perplexity AI

The good news is that almost none of this requires rebuilding your site. It's technical hygiene with a new priority.

Serve content in the HTML, not just on the client

If your page renders everything in the browser, do an honest test: disable JavaScript and read the result. What remains is roughly what the agent sees. Price, delivery time, coverage area, and differentiators need to be in the served HTML. In Next.js, this means preferring server-side rendering for the content that matters and leaving the client for interaction.

Write direct answer blocks

Before developing a topic, answer the question in two or three self-contained sentences, with numbers and dates. Such a block can be extracted whole. A paragraph that only makes sense after reading the previous three cannot. Put the summary at the top of the section, not at the end.

Mark facts with structured data

Organization, LocalBusiness, Product, FAQPage, and Article still matter — and now they also serve to disambiguate your brand in generated answers. Include your Companies House number when relevant, address, phone, author with real credentials, and a visible update date. Consistency between your site, Google Business Profile, and social media reduces the chance of the AI confusing you with a namesake.

Define robots.txt and consider an llms.txt

Decide consciously which AI crawlers you accept. If the goal is to appear in answers, blocking the crawler is shooting yourself in the foot. The llms.txt, a markdown file at the root pointing to your canonical pages, is still an informal convention and doesn't replace anything — but it costs an hour and organises what you want read first.

How to measure traffic and citations from Perplexity AI

Without measurement, this becomes an opinion debate. Three layers solve it.

Referral in analytics. Create a segment with AI origin domains (perplexity.ai, chatgpt.com, gemini.google.com, copilot.microsoft.com) and track sessions, landing pages, and conversion separately from organic. You'll find that volume is small and conversion rate is high.
Server logs. Filter by AI crawler user-agent. What matters is crawl frequency, which URLs are requested, and how many 404s or 403s you're serving by mistake.
Manual citation audit. Once a month, ask the ten questions your customer would ask before buying, in the product. Note who was cited. It's laborious and it's the only reliable way today to know if you exist in this channel.

Anyone who already tracks paid traffic panels will recognise the logic: new channel, low volume, high quality, and the right decision is to isolate the measurement rather than dilute it in organic. It's the same discipline we apply when analysing Google I/O 2026 announcements for UK businesses.

Five pitfalls to avoid

Trying to trick the agent. The Time case shows the cost: blocking and loss of trust. Manipulating content for bots is the new cloaking.
Blocking all AI crawlers out of fear. Without crawling there's no citation, and Perplexity AI only cites what it can read. Block what's sensitive, allow what's commercial.
Treating AI as just another source of clicks. Much of the value is brand mention without a click. If your goal is only sessions, you'll declare failure too early.
Publishing content without a date or author. Models that verify sources discard material that can't be dated or attributed.
Ignoring the post-click experience. Being cited is of little use if the person arrives at the site and can't find an obvious way to talk to someone.

From click to service: what to do with people arriving via AI

People arriving via Perplexity AI arrive differently. They've already compared, they have technical vocabulary, and they usually ask something specific right away. If the only door is a generic contact form with a response within 48 hours, you lose the advantage the citation gave you.

The pattern we use in Agathas Web projects is simple: a page with the explicit fact, a direct answer block at the top, and an immediate conversation channel — usually WhatsApp with the Official API, so the first contact is answered in seconds and the history isn't lost when the conversation passes to a human. It's not about a bot answering everything; it's about not letting a lead that came ready go cold.

If you want to understand better how this type of agent is being packaged by the big tech companies, it's worth reading Google Gemini: What Changed at I/O 2026.

Conclusion: start with what you can do this week

Perplexity AI won't replace Google in the UK in 2026, and it doesn't need to. It's already the second AI search channel in the country, with a visitor profile that converts above average, and it just gained legal backing for the type of browsing that will generate more agent visits to your server.

My priority roadmap for the next seven days, in order: disable JavaScript and see what remains of your three highest-revenue pages; rewrite the top of those pages as direct answer blocks with numbers and dates; check robots.txt to make sure you're not blocking who you want to attract; and create the AI referral segment in analytics to have a baseline.

These are four half-day tasks that put you ahead of most competitors, who are still debating whether AI is a fad. If you need help auditing your site from this perspective, it's exactly the kind of work we do here.