AI personal assistants: what Instinct, Dots, Muse and Spark mean for your website

The short answer

Instinct raised $1 billion at a $10 billion valuation on 28 September 2026, without disclosing a single user number (Business Wire). Cloudflare reports that daily requests from AI agents on its network grew by more than 1,700% in a year (Cloudflare).

What is an AI personal assistant?

An AI personal assistant is software that carries out tasks for one person rather than just answering questions. It connects to their email, calendar and accounts, runs on a computer in the cloud, and keeps working after the chat window closes. It books, compares, fills in forms and buys, and it asks before the risky steps.

A chatbot sends someone to your website. An assistant visits it for them and reports back.

Which AI personal assistants launched in 2026?

Four launched in five months, and they share one design: a computer in the cloud, connections to the user's apps, and work done without the person watching. Here is what each maker says, checked at its own source on 11 October 2026.

Gemini Spark

Google calls Spark "a 24/7 personal AI agent" that runs on dedicated machines in Google Cloud, so the user's laptop can be closed (Google). It is built into Workspace, and Google named Canva, OpenTable and Instacart as its first partner app connections. Google Australia announced it here in July for AI Ultra and Pro subscribers, and says it asks first before high-stakes actions "like spending money or sending emails" (Google Australia). Of the four, it is the only one with an Australian launch announcement.

Instinct

Instinct is a startup assistant you text or call. Its website says it uses a phone and computer "in the same way that humans do" (Instinct). Its Concierge tier makes phone calls to businesses without online booking. Its privacy policy says users may hand over usernames and passwords so it can sign into accounts on their behalf (Instinct privacy policy). TechCrunch reports it opened as an invite-only service in August (TechCrunch).

Meta Muse

Meta launched Muse in the United States on 8 September. Its own page says Muse "has access to a full web browser, so it can search, navigate sites, fill out forms, and complete transactions" (Meta). Meta's security write-up adds the detail that matters to site owners: the browser sub-agent reads "an accessibility tree snapshot of the page", not the raw code, and when Muse browses, "it will appear as your activity" (Meta AI research).

OpenAI Dots

OpenAI launched Dots at DevDay on 29 September, describing them as "always-on agents built to handle everything" (TechCrunch). OpenAI's announcement says each dot runs on its own cloud computer with its own browser, connects to over 4,000 apps through plugins, and uses only read-only tools when it researches on its own initiative, so in that mode it cannot send messages or control the user's browser (OpenAI). The first dot comes with ChatGPT Pro and Business Premium.

How do AI personal assistants reach your website?

They reach it in two ways. Where your business has an app integration or API, the assistant uses that. Where it does not, which is most small and mid-sized business websites, the assistant opens your pages in its own browser, reads them, and clicks and types like a person. Your website is the fallback interface, and for most businesses it is the only one.

That has three consequences.

• The assistant reads structure, not design. Meta's browser agent works from the accessibility tree: the names, labels and roles of what is on the page. A button with no label, or a price shown only in an image, is invisible to it.
• Speed and stability count. An assistant that waits on a slow page or misses a moved button tries the next supplier.
• Not every fetch runs your code. Browser agents run JavaScript. Many lighter fetchers that assistants use to look things up do not. A page that builds its words with scripts can read as empty to them. Our free tool can AI see your page shows you what a non-browsing fetcher receives.

What did Amazon's block on Meta Muse show?

It showed that letting assistants in is a business decision. On 20 September Amazon said it had blocked Muse from shopping on Amazon.com after asking Meta to leave Amazon out of the product. Muse users now see a warning that continued access by an unauthorised AI agent breaches Amazon's Conditions of Use, according to GeekWire.

Amazon's reasons, as reported by GeekWire: it never agreed to Muse using its store, Meta did not tell it in advance, Muse does not identify itself when it browses, and Muse appears to capture and store customer credentials. Meta's launch materials say Muse has no visibility of passwords or payment methods, because credentials are stored outside the model. The two positions differ in emphasis more than in fact.

Amazon's own Conditions of Use, updated on 14 August 2026, already set the rule. They require any agent to identify itself at all times:

The same section bans agents from mimicking human keystrokes, getting around CAPTCHAs or evading blocks, and reserves Amazon's right to limit agents "at our sole discretion" (Amazon).

Our view: Amazon can afford to say no because people will open Amazon themselves. Most businesses cannot. A plumber, a law firm or a software company blocking assistants is not protecting a marketplace. It is removing itself from the shortlist an assistant hands back to a buyer. Decide deliberately, and for most businesses the decision is to let identified assistants in.

Can you tell when an assistant visits your site?

Only partly. We checked each maker's own documentation on 11 October 2026 for how its assistant identifies itself to websites. The answer is uneven, and it explains why Amazon reached for its terms rather than its robots.txt file.

Google, OpenAI and Meta all say their user-requested fetchers may not follow robots.txt, because a person asked for the page. Cloudflare says OpenAI, Google and AWS now sign agent requests with Web Bot Auth (Cloudflare). Identification is improving, but an assistant that browses as its user is, to your server, just a visitor.

Our view: robots.txt still matters for crawlers, so check it. Our guide to ClaudeBot and Anthropic's three crawlers explains the training, search and user-request split that every maker now follows, and the AI crawler access checker reads your file for 20 named bots. But the bigger risk for most businesses is the firewall or bot-protection setting that challenges anything unusual. An assistant that meets a CAPTCHA is not allowed to solve it, so it leaves. We hold ourselves to the same rule: PostCliq's own crawler names itself and explains what it does on our bot page.

What does the second audience need from your website?

It needs what a careful buyer needs, written down rather than implied. An assistant is a buyer's representative on an errand: find three suppliers, compare price and availability, book or enquire. It reads text, labels and structured data, and it needs a way to finish the job or hand it back.

Clear prices and policies win the comparison

An assistant asked to "find a physio near me who takes my insurance and has a Saturday slot" will compare the facts it can find. If your fees or hours sit in a PDF or behind "contact us for pricing", it has nothing to compare and moves on. Ranges are fine. Silence loses.

Forms and contact paths must work for a machine and for the person

The three big-company assistants all pause before high-stakes actions. Spark asks before spending money or sending email. Dots keep sensitive tasks such as password changes with the person. Muse stops for anything "hard to undo". So an errand often ends in a handover: a filled form awaiting approval, or a phone number passed back. A form with unlabelled fields, a script-only submit button or a CAPTCHA on every enquiry fails that handover. Our guide to contact form best practices covers the human side, and the same fixes serve the assistant.

Our view: why PostCliq is betting on the agentic web

PostCliq is a post-click optimisation platform. We find where websites lose enquiries and sales after the click. The assistant wave is the reason we think post-click work matters more, not less.

Here is the forecast, labelled as ours. People will browse less. Assistants will do more of the searching, reading and comparing, and they will make more of the shortlist decisions on their user's behalf. The human click will not disappear. It will arrive later, from a person who has already been briefed by an assistant, and it will carry more intent. Adobe's August 2026 data points the same way: visitors referred by AI to US retail sites converted at a 60% higher rate than other traffic in July (Adobe Digital Insights). That is US retail and AI referrals, not assistants alone.

So the post-click journey now has two readers. The assistant must be able to get in, read the facts and reach the form. The person who follows must land on a page that confirms what they were told and lets them act. A site that fails either loses the enquiry. That is post-click optimisation in this era, and it is where generative engine optimisation and getting cited by ChatGPT were already pointing towards.

What we do not know is how fast this happens. None of these products is a year old, and Instinct has published no user numbers. Build for what they share, not for any one of them.

What to do next: a checklist for this week

1. Score your site for agents. Run the free agent readiness check on your homepage.
2. Decide your access policy. Ask whoever manages your hosting, CDN or security plugin whether bot protection challenges automated visitors. If it does, allow identified agents and user-requested fetchers such as Google-Agent, ChatGPT-User and Meta-ExternalFetcher, and check the rest with the AI crawler access checker.
3. Check your money pages load as text. Use can AI see your page on your top service, pricing and contact pages. If the words appear only after JavaScript runs, ask your developer for server rendering.
4. Write the facts down. Put prices or price ranges, service areas, hours, delivery times, returns and booking rules on the page as text, and match them in structured data with the schema markup checker.
5. Label every form field and button. Each needs a visible label and an accessible name. Then submit your own form and check the enquiry arrives, using can visitors contact you.
6. Remove CAPTCHAs from the first step. Use quieter spam protection, so an assistant can complete or hand back the enquiry.
7. Give a human fallback. Put a phone number, email address and booking link in plain text on every contact and service page, so an assistant can pass them to its user.

For the whole post-click journey checked at once, run a free PostCliq audit. More on this shift lives in our AI search and visibility articles.

Frequently asked questions

What is Instinct AI?

Instinct is a personal AI assistant made by Spear Street Technology, a San Francisco startup founded by Noah Shinn. You text or call it, and it completes tasks such as bookings, grocery orders and bill disputes using its own phone and computer. It is in early access and raised $1 billion at a $10 billion valuation in September 2026.

What are OpenAI Dots?

Dots are always-on AI agents in ChatGPT, launched by OpenAI on 29 September 2026. Each dot has its own cloud computer and browser, connects to more than 4,000 apps, and works on your goals in the background, with permissions and approvals controlling what it may do. The first dot is included with ChatGPT Pro and Business Premium in eligible markets.

Why did Amazon block Meta Muse?

Amazon said Meta never asked permission for Muse to shop on Amazon.com, that Muse does not identify itself when it browses, and that it appears to store customer credentials. Amazon's Conditions of Use require agents to name themselves in every request. Meta says Muse cannot see passwords or payment methods. The block took effect in September 2026.

Should my website block AI personal assistants?

For most businesses, no. Blocking an assistant removes you from the shortlist it gives a buyer, and unlike Amazon, most businesses do not have customers who will come back directly. Allow identified assistants, keep blocking abusive traffic, and make sure the pages they read carry your facts and a contact path they can complete.