Search Console has finally started to answer two questions marketers have been asking for a couple of years: how often does Google show our pages inside AI answers, and how much of our traffic starts with a photo rather than a typed query? Over the past few months Google has released a generative AI performance report, and on 24 September 2026 it began rolling out a new multimodal filter in the main Performance report.
This guide explains what the Search Console AI performance data actually contains, what the multimodal filter separates out, and what neither of them can tell you. It then walks through a reporting workflow you can set up this month, with the caveats you should write into any client or board report.
What Google has released, and when
There are several moving parts, so it helps to separate them. Here is the timeline as Google and the trade press have reported it.
- 3 June 2026: Google announced Search generative AI performance reports in Search Console, together with a control that lets a site opt out of generative AI features. Testing started with a limited group of site owners in the UK.
- 31 August 2026: Google’s note says the insights have been rolled out to all websites worldwide. Its own help page adds that sites with too few generative AI impressions will not see a report at all.
- 2 September 2026 (approximately): Google Trends added category filtering on its Explore page, which is useful for the research side of this story.
- 16 September 2026: John Mueller said position and impression reporting for AI Mode, AI Overviews and classic Search will evolve over time, because those products evolve too.
- 24 September 2026: Google started rolling out the web multimodal filter globally. It appears in the Search results report and in the generative AI report, and only for sites that receive this kind of traffic.
Two cautions before we go further. First, Google has said the multimodal rollout is gradual, so you may not see the filter in every property yet. Second, several details below are things Google has not published, and I will flag those as we go rather than guess. For the official wording, read Google’s help page for the generative AI performance report and its announcement of web multimodal reporting.
How the generative AI performance report works
The report counts impressions: how many times a link to your site was shown to someone inside a generative AI feature on Google Search. At the time of writing, Google’s help page names AI Overviews and AI Mode, and says the list may grow. Search Labs experiments are excluded. Discover also has its own generative AI view, reported separately from Search.
You can break the data down in four ways:
- Pages: the final URL after redirects, grouped under its canonical URL. This is the most useful tab, because it shows which of your pages Google chooses to surface in AI answers.
- Countries: where the search came from.
- Devices: desktop, tablet or mobile (on the Search side).
- Dates: by day, week or month, in Pacific Time. Your own analytics probably uses a different time zone, so expect small mismatches when you compare days.
What is not in the report
This matters as much as what is. Google’s help page describes only impressions. It does not list clicks, click-through rate or average position for the AI view, and queries are not available either. Coverage of the launch also notes that the report counts links to your site only: it does not tell you whether your brand was mentioned in an answer without a link, which competitors appeared beside you, or whether the answer was favourable.
In practice, you can see that a page was shown, but not what the person asked, and not whether they clicked. That is still progress. Until now most teams relied on third-party prompt trackers, which sample a fixed set of questions. This is Google’s own count of real exposure. Our guide to automating AI search visibility tracking covers how to combine the two sources.
It is not a separate pile of traffic
The generative AI impressions already sit inside the Web search type of the main Performance report. The new view is a slice of data you were already counting, not an addition to it. If you add the AI report to your normal Search totals, you will double count. Treat it as a “how much of my visibility happens inside AI features” lens.
The multimodal filter explained
Under the Search type control, the Web option now splits into Text-based and Multimodal. Google’s wording for multimodal is web search results where an image was used as part of the search. The entry points Google names are:
- Google Lens: pointing a phone camera at an object or uploading a photo.
- Circle to Search on Android: circling or tapping something on screen to search for it.
- Image uploads to Google Search: dragging in or choosing a photo.
- Chrome’s “Search this image” option: right-clicking an image on a web page.
Google has not said whether other features, such as image prompts in AI Mode or Search Live, count as multimodal, so do not assume they do. It has also not explained exactly how a search that mixes a photo with typed words is classified.
What you get, and what you do not
With the filter on, you can see which of your pages appeared for image-led searches, and compare them with text-based traffic by country and device. As John Mueller put it, you can see which pages showed up and how often. What you cannot see is the query: the Queries tab is unavailable, because Google says specific text query data does not exist for this traffic. You will not see the photo either.
That changes how you interpret it. This is a page-level and product-level signal, not a keyword report. You learn which pages, products or guides are being matched to images, and you infer intent from the page itself.
Why it matters now
Image-led search is not a niche behaviour any more, and our overview of voice and multimodal search in 2026 explains the user side in more detail. Until this filter, none of that activity had a home in your reporting. Google has said the data was not previously included in the Web counts, which Mueller confirmed, so some properties may see new numbers appear rather than a re-sorting of old ones.
How to read the numbers without fooling yourself
The new reports are easy to misread, and the risk is higher because the numbers will arrive in a meeting before anyone has checked them. These are the points to keep in mind.
- Impressions are not visits. An AI impression means a link was shown. With no click data in the AI report, it says nothing about traffic. Do not present it as such.
- Position is unsettled. Mueller said on 16 September that AI position and impression reporting will change over time and that there are edge cases. He described the aim as being useful to site owners rather than an absolute truth for position counting. He also warned that site owners will need to put in work to interpret and compare the figures. Earlier reporting suggested that AI Overview links were all given position one while AI Mode positions follow placement, but that is not something Google has restated in its new documentation, so treat any AI position figure as directional.
- Totals may shift. One SEO professional reported drops in Web impressions and clicks on some properties from around 10 September and linked them to the new data. Mueller said he would check, and no explanation had been published when I looked. If your Web totals moved in September, annotate the chart rather than explain it away.
- Row limits apply. The usual 1,000-row limit on tables still applies, so filter by country, device or URL folder on large sites.
- The newest days are preliminary. They show as a dotted line. Do not report on yesterday.
- Exports clean up values. Google notes that exporting converts “~” or “-” values to zeros, so keep the on-screen view for anything that looks suspicious.
One more thing we could not confirm: whether the multimodal data is available through the Search Console API or bulk export. Google’s post does not mention it, and the API’s search type parameter does not list a separate multimodal value. Until Google says otherwise, plan on manual exports for this slice.
Building a reporting workflow that holds up
You do not need a new tool. You need a repeatable routine and a few clean definitions. Here is a set of steps we would use for a client account.
- Check access first. In each property, open the Performance report for Search results and look for the Search type filter. Then look for the generative AI report. If either is missing, note which properties lack it. It may be a rollout delay, or the site may simply have too few AI impressions.
- Set a baseline. Export the full available date range for AI impressions by page, country and device. Do the same for the Web multimodal filter. Save these as dated files, because the data will keep changing as Google refines it.
- Tag the pages by type. Group URLs into product pages, category pages, guides, location pages and everything else. Page type is the only lens you have, since queries are missing.
- Compare AI exposure with clicks. For each group, put AI impressions next to clicks from the normal Web report. Pages with plenty of AI exposure and weak click growth are worth examining: perhaps the AI answer is satisfying the question, perhaps the snippet is weak.
- Review the multimodal pages. Which pages appear for image-led searches? Look at their images, alt text, product details and structured data. We cover the markup side in our guide to schema markup for AI search visibility.
- Log changes. Keep an annotation sheet of site releases, content updates and Google announcements. With the data in flux, annotations are what let you explain a movement later.
- Report monthly, review weekly. Weekly checks catch problems; monthly reports avoid overreacting to noise.
If you already pipe Search Console into a warehouse, add the new exports alongside the old tables rather than overwriting anything. Our walkthrough of Search Console and BigQuery reporting shows the structure, and our notes on automating daily SEO tasks cover how to schedule the routine checks.
Using the data to make content decisions
Reports only earn their place if they change what you do. Here are practical uses for each dataset.
From the AI performance report
- Find your “AI-friendly” pages. Pages that appear often in AI features show what Google already trusts you for. Study their structure: clear definitions, tidy headings, direct answers, current facts.
- Spot the gaps. If a page type you care about, such as comparison guides, rarely appears while thinner pages do, that points to a content depth problem.
- Prioritise refreshes. A page with healthy AI exposure but old facts is a risk. Fix those first. Our piece on automated content refreshes and content decay describes a workable cadence.
- Check the zero-click effect. If AI impressions rise while clicks stay flat, you are likely seeing the pattern described in our zero-click search statistics article. Adjust your goals toward brand search, leads and assisted conversions instead of sessions alone.
From the multimodal filter
- Retail and local businesses: check whether product and location pages are the ones being matched. If a bestseller is missing, review its photography: clean backgrounds, several angles, descriptive file names and alt text.
- Publishers and service businesses: image-led searches often come from people identifying something, such as a plant, a part, a dish or a landmark. Pages that name and explain things clearly tend to suit that need, though Google has not published guidance on this, so test rather than assume.
- Brands with a visual identity: if screenshots of your product or ads are shared widely, a multimodal search may land people on your pages. Make sure those pages work on mobile, because Circle to Search is an Android feature.
A note on Google Trends
The category filter added to the Trends Explore page in early September helps you narrow results to a topic area, and, according to Google, you can browse top-searched terms by category, region and timeframe without entering a keyword. It will not show you your own data, but it is a useful way to check whether a rise in AI or multimodal exposure reflects wider interest. For a broader view of who is winning AI-driven attention, see our AI search market share analysis.
Common mistakes to avoid
- Treating AI impressions as a ranking. They are exposure counts, and the position side is still being worked out.
- Adding AI impressions on top of Web impressions. They are already inside the Web total.
- Promising clients AI traffic numbers. No click data exists in the AI report, so any traffic estimate is your own inference.
- Hunting for multimodal keywords. There are none. Use pages, countries and devices.
- Ignoring the September timing. New data entered the reports around the same time as other changes. If you see a break in the trend, annotate it before drawing conclusions.
- Confusing the opt-out control with an optimisation lever. The generative AI control is a property-level setting that excludes your content from the AI features. Opted-out sites get no impressions or traffic from them. Google says it is not used as a ranking signal elsewhere in Search and does not affect AI training, which is governed separately by Google-Extended. It is a business decision, not a tactic, and it should involve more than the SEO team.
- Relying on one source. Search Console shows Google’s view only. Other AI assistants are outside it, so keep your wider tracking in place.
- Reporting without caveats. A chart with no notes about limits invites bad decisions. Add a short methodology box to every report.
If your team uses AI assistants to pull this data, our guide to MCP for marketers explains how to connect them to Search Console safely, and why you should still check the output against the interface.
A practical action plan
Here is a sensible order of work for the next 30 days.
This week
- Confirm that the generative AI report and the multimodal filter are visible in each property you manage, and note any that are not.
- Export baseline data for both, and save it with a date.
- Annotate September in your reporting, including the 24 September rollout.
Weeks two and three
- Group your top pages by type and compare AI impressions with clicks.
- List the pages that appear in multimodal results and review their images, alt text and product information.
- Pick five pages with strong AI exposure and out-of-date content, and refresh them.
Week four
- Produce a one-page summary for stakeholders: what we can see, what we cannot, what we changed and what we will watch next.
- Decide who owns the monthly export and when it happens.
- Revisit Google’s help pages for any documentation changes, since Mueller has said it will be updated if there are significant changes.
The data is new, incomplete and still moving. That is a reason to start collecting it now, with clear caveats, rather than to wait for a perfect report that Google may never publish.
Frequently asked questions
What is the multimodal filter in Google Search Console?
It is a new option under the Web search type in the Performance report. It separates searches where an image was part of the query, such as Google Lens, Circle to Search, image uploads and Chrome’s Search this image, from text-based searches. Google began rolling it out globally on 24 September 2026, and it appears only for sites that receive this traffic.
Can I see the queries behind multimodal searches?
No. When the multimodal filter is selected, the Queries tab is unavailable, because Google says specific text query data is not available for this traffic. You can see which pages appeared and how they perform by country and device, so you infer intent from the page rather than a keyword.
Does the AI performance report show clicks or position?
Google’s help page for the generative AI performance report describes impressions only, for AI Overviews and AI Mode, broken down by page, country, device and date. It does not list clicks, click-through rate or average position. John Mueller has said AI position reporting will evolve over time.
Why can't I see the generative AI report in my property?
Google says the insights reached all websites worldwide on 31 August 2026, but sites with too few generative AI impressions will not see a report. Sites that have opted out of generative AI features will not see one either. Some properties may also see the multimodal filter later than others during rollout.
Do I add AI impressions to my normal Search impressions?
No. Generative AI impressions are already counted within the Web search type of the main Performance report. The AI view is a slice of existing data, so adding it to your total would double count. Use it to see what share of your visibility happens inside AI features.
Is the multimodal data available in the Search Console API?
Google’s announcement does not say, and the API’s search type parameter does not list a separate multimodal value at the time of writing. The report has an Export button, so use manual exports for now and check Google’s API documentation for changes before building automated pipelines around it.
Our team builds reporting workflows that turn new Google data into clear, practical content decisions.
Riya Bhardwaj
Leading content and growth initiatives with a focus on search visibility, audience engagement, and measurable business outcomes. Specialised in SEO, Generative Engine Optimization (GEO), AI search optimisation, and performance-driven content marketing. Passionate about transforming market insights into scalable content strategies that strengthen brand authority and drive sustainable digital growth.