Lesson 2: Confirmed, Test or Rumour: Verifying AI Channel Claims

Why this matters

In Level 1 you saw how people now research and buy through AI assistants. The channels inside those assistants change every month. A feature that was a small test in spring can be live in 60 countries by autumn, or it can be shut down. If you plan budget around a blog post's claim, you may be planning for something that does not exist in your market.

This level gives you one method you can repeat. With it you sort any AI marketing claim into one of three groups and write down the evidence.

The three statuses

• Confirmed: The vendor's own documentation, help centre or official announcement says the feature is live. It also names who can use it and where.
• Test / preview: The vendor confirms the feature exists but calls it a test, beta, pilot, early access, "early stages" or limited rollout. It may be restricted to certain countries, languages, account types or invited advertisers.
• Rumour: Only unofficial sources mention it, such as trade press, agency blogs, social posts or leaked screenshots. The vendor has not confirmed it, or the vendor's current pages contradict it.

A useful comparison is a planning application. A neighbour saying "they're building flats there" is a rumour. A filed application is a test or preview. Approved permission with conditions attached is confirmed. Even when something is confirmed, you still have to read the conditions.

Task 1: Classify a claim

DO this for each claim you come across:

1. Write the claim in one sentence. For example: "Ads now appear inside Google AI Overviews in the UK."
2. Find the primary source. Go to the vendor's help centre, developer docs, official blog or newsroom, or the advertiser platform's availability page. Do not stop at an article that only describes what the vendor said.
3. Look for status words. "Testing", "beta", "preview", "pilot", "early stages" and "select advertisers" all mean test/preview, even if the press calls the feature "launched".
4. Check scope. Note the countries, languages, devices, plans and account types listed.
5. Assign a status. It is confirmed only if the vendor says the feature is live and your market is in scope. Otherwise it is test/preview (for someone else), or rumour.

Worked example (read 20 October 2026): Google's Ads Help page "About ads and AI Overviews" says ads within AI Overviews are available in English in 12 countries: Australia, Canada, India, Indonesia, Kenya, Malaysia, New Zealand, Nigeria, Pakistan, Philippines, Singapore and the US. It also says the feature is "still in early stages". The UK is not on that list. So for a UK business, the claim above is not confirmed for your market.

Task 2: Record it in a channel tracker

A status without a date goes stale quickly. DO set up a simple spreadsheet with these columns:

• Channel / assistant (e.g. ChatGPT, Google AI Overviews)
• Feature (e.g. ads below responses)
• Status: Confirmed / Test-preview / Rumour
• Source URL (the primary source, not a news article)
• Date read (the day you checked, not the article's publication date)
• Markets (countries and languages)
• Eligibility (plans, account types, verticals excluded, invite-only?)
• Next review date

Example row: ChatGPT | Ads | Test began 9 February 2026 for some adult Free and Go users in the US | source URL | date read | markets | eligibility. Reports this autumn give different country counts, including "over 40" and "over 60". When figures disagree like this, do not copy any of them. Go to the vendor's own availability page and record what it says on the day you read it.

DO set a review cycle. Monthly suits most channels. Check fortnightly for any channel you are actively spending on.

Task 3: Spot warning signs in unofficial reports

DO treat a report with caution when you see:

• Unsourced figures: performance claims such as "2x CTR" or "CPMs of £X" with no link to vendor data or a named study.
• Out-of-date dates: the article is months old, or calls something "coming soon" when that date has already passed.
• Retired or relaunched claims: features that have been paused, ended or rebranded. Perplexity is an example. It tested sponsored follow-up questions from 2024, and in February 2026 it was widely reported to have moved away from ads. Reports differ on when the ads actually stopped, and they cite secondary sources such as the Financial Times. Record the conflict and look for Perplexity's own statement.
• Disagreeing numbers: different sites quoting different country counts or launch dates.
• "Launched" with no scope: no countries, languages or eligibility given.
• Screenshots only: one user's interface is not proof of general availability.
• Vendor-selling sources: agencies or tool vendors who benefit if you believe a channel is open.

If you see one or more of these warning signs, keep the claim at Rumour until a primary source settles it.

Readiness task

1. List five AI channel claims you have heard in the last month.
2. Classify each one using Task 1.
3. Add each to your tracker with its source URL and date read.
4. Flag any claim that has a warning sign from Task 3.
5. Set a review date for each row.

Checklist

• Every claim has a primary source URL
• Every row has a date read
• Markets and eligibility are filled in
• Status words (test, beta, early stages) are reflected in the status
• Claims with conflicting reports are marked as such
• Review dates are set

Key takeaways

• Use three statuses: confirmed, test/preview and rumour.
• Only the vendor's own documentation or announcement can confirm a feature.
• "Confirmed" only counts if your market and account type are in scope.
• Always record the source URL and the date you read it.
• Unsourced figures, stale dates, disagreeing numbers and retirement claims are warning signs.