#BlinkTank: Shopify's data shows AI shoppers convert twice as well
Shopify released a set of its own Q2 commerce data yesterday, and it makes a strong case for something we say often: fix your product data at source and both AI and organic search take care of themselves. According to Shopify, AI-referred shoppers are converting up to twice as well as organic shoppers in some categories, and much of this is down to getting the fundamentals of product data right,
After that: a wave of new AI features Google has just pushed into Ads and Analytics, then the smaller stuff worth being aware of - a possible GA4 bug over-inflating Direct revenue, a noisy few days in Google's rankings, a useful new split in how Microsoft Clarity reports AI citations, and a Google Ads deadline on 1 September. We finish on Bending Spoons buying Airtable, which matters more to us than it might first look.
On to the news!
Shopify's own data shows AI shoppers convert twice as well
Shopify published a set of Q2 commerce data yesterday looking at how AI-referred and organic-referred shoppers behave differently, and a lot of it validates what we talk about here week in, week out.
The summary: AI-referred sessions to Shopify stores grew 197% year on year in Q2, roughly three times, with orders growing at about the same rate. Organic search sessions grew 12% over the same period, from a much higher base.
The more interesting number is where AI converts best. Across the board, AI-referred shoppers converted around 80% better than those from organic search, and that gap widens in categories where people compare specs before buying - close to double the conversion rate in categories like watches or necklace.
What we liked most is how clearly Shopify makes the point that optimising for AI and optimising for organic come from the same place - fixing your product data at source. Structured Shopify Catalog data converted AI-referred shoppers at roughly double the rate of scraped or third-party feed data, and Shopify's own recommendation is to add product attributes that support conversational discovery - writing product content in the language shoppers actually use to ask questions, not just the language merchants use to classify SKUs internally. That is the same argument we keep making about clean product data working for both organic and AI.
Treat it as Shopify's own data, published in part to promote its Agentic feature and Shopify Catalog. But none of this surprises us, and it is, frankly, music to our ears.
🔗 AI and organic search are doing different jobs (Shopify)
Shopify keeps investing in files for agents
Cast your mind back a few months and you will remember the wave of interest when Shopify switched on a set of agent-facing files across every store: /llms.txt, /llms-full.txt, /agents.md and a /sitemap_agentic_discovery.xml tying them together. We talked about it a lot, as we do about this sort of thing.
At the time everyone had a look, came up with theories about what it meant, and then pretty much stopped caring, mostly because it became clear the files were not doing very much.
Shopify has not given up on them, though - over the past few weeks it has reworked the whole setup again.
The three text files now carry identical content. That is more of a shift than it sounds, because /agents.md was the transaction-focused interface and /llms.txt the discovery layer, with /llms-full.txt a longer version of the latter, and merging them collapses a distinction Shopify had been careful to keep.
The new content is written far more like a natural-language prompt than a technical spec, and it steers AI platforms towards the shop.app/SKILL.md file for the actual transacting. It is worth remembering /agents.md is not a Shopify invention but a Linux Foundation open standard already used by tens of thousands of projects, originally for coding agents, which Shopify has adopted as its handshake. The agentic discovery sitemap that used to reference all of this has been condensed down to just /agents.md.
As I mentioned above, many in the industry have moved on from things like llms.txt. In June, Ahrefs released an excellent study that found that 97% of llms.txt files are never read by AI platforms.
As it stands, if you're looking for work that drives tangible growth right now, there's probably better ways to spend your time. The key phrase, though, is right now. Ahrefs study is a snapshot of a something still being built out.
This is the thread running through the last few months of BlinkTank. Shopify is investing in the infrastructure well ahead of the demand, in the same way it has with UCP and agent checkout through Shop Pay. I would be extremely surprised if it were pouring this much effort into something it expected to go nowhere - far more likely, it sees these files becoming useful and wants the plumbing in place before that happens.
🔗 Ahrefs study on whether llms.txt is used
Google adds AI dashboards and benchmarking to Ads and Analytics
Google has started rolling out a set of AI features across Google Ads and Google Analytics, all built on its Ask Advisor agent. There are three parts to it. The homepages of both tools now open with AI-generated overview cards summarising what has changed since you last logged in, with a click carrying a card into Ask Advisor to look closer, and an option to receive the summaries by email or phone. Google Ads has gained Dashboards, coming to Analytics later, which build visual reports from a plain text prompt and attach a written summary explaining the movement. And Analytics has added benchmarking, comparing your campaign performance against anonymised averages from similar businesses.
For a large-catalogue store the reporting side is the part most likely to be of interest. Describing the report you want in a sentence and getting a chart plus a first-pass explanation of the "why" sounds like a great feature. However, as ever, the devil is in the detail.
The benchmarking is the part to approach with more caution. We have not dug into Google's version yet, but the recurring weakness in benchmarking is the peer set it puts you in.
A comparison is only really meaningful when it matches you on the right dimensions. Over the years, we found that things like platform, revenue bracket, catalogue size and vertical are the ones that really matter.
For example, a Shopify store turning over £4-5 million a year with 5,000 products, compared against other stores that meet the same profile, is something you can trust. If you group on vertical alone though, as these tools tend to, it falls apart - a single-SKU dog food brand and Pets at Home are in the same sector, yet almost nothing about their conversion rate, traffic mix or average order value is comparable. Until a benchmark can match on that fuller profile, an anonymised average from your "similar businesses" is often useless.
Back to these updates, there are two caveats worth keeping in mind. It is in beta and limited to English-language accounts for now, so it may not have reached your account yet, and the summaries are Google's AI describing your own data back to you with no context, so please exercise judgement.
🔗 Evolve your marketing with new AI tools (Google)
A possible GA4 attribution bug is inflating Direct revenue
This is one to check on your own GA4. Elvinas Karalis at Digital Bees has flagged that, since 2 August, Direct has been taking a growing share of purchase revenue across several unrelated GA4 properties, with Paid and Organic losing revenue at roughly the same moment. Sessions stay normal, so it is the revenue attribution moving rather than the traffic itself. The pattern so far is specific: only properties using the Blended reporting identity, and only EMEA clients, with US and Asian properties apparently unaffected.
We checked it across our own client base, and only a handful of stores showed the pattern, so it is not a widespread problem, and on the accounts where it did appear it seems to have settled back to normal from around 7 August. The reason to keep an eye on it is that attribution glitches like this have a way of causing knock-on problems, in reporting or in anything that feeds off GA4 channel data.
The practical thing now is to look at your own Direct revenue trend from 2 August before it turns up in a report and someone asks why Direct suddenly looked like your best-performing channel.
🔗 Elvinas Karalis on the GA4 Direct revenue anomaly (LinkedIn)
Ranking volatility at the start of August
The first week of the month brought a stretch of unstable rankings. Search Engine Roundtable flagged heightened volatility between 1 and 3 August, with third-party tracking tools spiking, and the movement seemed to carry into the 5th and 6th. Plenty of site owners reported sudden shifts across Search and Discover over the same days.
So far Google has not confirmed anything. There is no entry on the Search Status Dashboard for a ranking, indexing or serving incident across those dates, and the reports look like a mix of Search, Discover and indexing wobble rather than one clean algorithm update. The last confirmed change is still the June 2026 spam update, which we covered when it rolled out. As ever the best response to unconfirmed volatility is usually to do very little - note where you moved, avoid sweeping changes, and wait to see whether the pattern continues.
🔗 Google search ranking volatility, 1 to 3 August (Search Engine Roundtable)
Microsoft Clarity splits AI citations into branded and non-branded
Microsoft Clarity has added branded query segmentation to its AI Citations dashboard. Credit to Aleyda Solís for flagging it in the SEOFOMO newsletter. Individual queries are now marked as branded or not, the Share of Authority card breaks the two out, and you can filter the whole dashboard by either, so you can see when an AI cited you because someone named your brand versus when it reached for your content to answer a general category question.
That split matters more than it first looks. Tracking AI citations as one number hides the thing you actually want to know. Branded citations tell you how often AI systems lean on you when the brand is already in the prompt, which is a measure of brand strength. Non-branded citations tell you how often your content gets picked for a broad, category-level question, which is where discovery happens and where a deep, well-structured catalogue should have the advantage.
🔗 Branded vs non-branded AI queries in Microsoft Clarity (Microsoft)
Google Ads auto-upgrades some Search campaigns to AI Max on 1 September
From 1 September, Google will start upgrading any Search campaign that uses automatically created assets, the campaign-level broad match setting, or both, to AI Max. Campaigns on automatically created assets move to AI Max with search term matching and text customisation on by default; campaigns using the broad match setting move across with search term matching on by default. This is separate from the target-based bidding change on 17 August we covered last month, so there are now two Google Ads shifts arriving within a fortnight of each other.
Our paid team has already moved our own campaigns to AI Max, less because we were sold on it than because the change is coming regardless and we would rather understand it on our own terms. The mechanics are clear enough: you give up some control over which queries you match and how your assets are written, and in return Google's automation broadens your reach through search term matching and generates more of the ad text and images itself. The automatic text and image customisation can still be turned off, which we expect to carry over after the upgrade, so it is not entirely all-or-nothing.
What is less clear is the benefit. Google's case is that the automation surfaces converting searches your keywords would never have caught, but on a well-run account it is hard to see what you gain that offsets the control you hand over. Either way, there is no clean way to leave things exactly as they are before 1 September: you either let the upgrade happen, turn off the legacy features beforehand to avoid it, or turn AI Max on yourself so you set the configuration rather than inheriting the defaults.
🔗 Google to auto-upgrade some Search campaigns to AI Max (Search Engine Land)
Bending Spoons agrees to buy Airtable
On 4 August, Bending Spoons announced a definitive agreement to acquire Airtable in an all-cash deal worth around $1.285 billion in enterprise value. It is the Italian company's first acquisition since it listed on Nasdaq on 1 July, and it joins a portfolio that already includes Evernote, WeTransfer, AOL and Eventbrite.
This one is close to home for us. We have been Airtable users for years and have built a huge amount around it, and beyond our own setup I have seen stores use it in very interesting ways. For a lot of large-catalogue brands it does a job Shopify's native fields cannot - acting as a lightweight product database that holds custom categories, supplier data and the messy attributes that never fit neatly into a metafield when you are dealing with thousands of SKUs. It is also the clean, linked, structured data that good AI work depends on, since a model can’t categorise a catalogue or draft accurate copy across thousands of products from records that are a mess.
When I first wrote about this on LinkedIn, I repeated the standard line on Bending Spoons: that it buys software, cuts costs, raises prices and slows the pace of new features. A few people have pushed back since, and fairly. Some users have seen the opposite, with the rate of releases picking up rather than slowing, so the reputation may be a little unfair. The more accurate description is that Bending Spoons tends to buy products that have gone stale and invest in turning them around.
That matters here because we think Airtable has a lot of untapped potential, particularly around AI, where clean structured data is the foundation everything else is built on. If new ownership accelerates that, it could be only be good news. Of course, none of that removes the sensible step of knowing what in your stack depends on Airtable and what it would take to move if the commercials change though.
🔗 Bending Spoons agrees to acquire Airtable (Bending Spoons)
Partner shout-out: AppStoreResearch
We have featured AppStoreResearch before, but it is well worth doing again. I had a great catch-up with Jonathan Kennedy the other day, and he remains one of my absolute favourite people in eCommerce.
If you build Shopify apps or commerce software, AppStoreResearch is well worth a look. It connects product and growth teams with vetted Shopify merchants for paid research interviews, handling the recruiting, screening, scheduling and incentives so you can talk to real operators about your roadmap rather than guessing from analytics. For merchants, it is the other side of the same coin - a way to join paid research calls and give app teams feedback before a tool reaches the app store.
🔗 AppStoreResearch
As ever, thanks for reading. If you found this useful, forwarding it on to someone else who runs or works on a large-catalogue Shopify store helps us more than just about anything.
See you next week.
Sam
Who am I and what does Blink do?
I'm Sam Wright, Managing Director and co-founder of Blink SEO. We're a search marketing agency that works exclusively with large-catalogue Shopify stores - the kind where the challenge isn't just visibility, but helping customers find the right product in the first place.
We specialise in SEO and PPC for merchants managing hundreds or thousands of products. Most of our work starts with taxonomy: how a store is structured and categorised, and what that means for organic performance, paid efficiency, and on-site discovery.
If you found this useful and aren't already subscribed, you can sign up at blinkseo.co.uk/blinktank.
If you run a large-catalogue Shopify store and want to talk about organic or paid performance, reply to this email or find me at blinkseo.co.uk.