Text or Multimodal Generative AI Impressions: Which Should You Act On First?

If Search Console now shows generative AI impressions under both Web: text-based and Web: multimodal, act on the search type that matches how buyers actually discover you—and fix different page signals for each. Text-based generative AI impressions usually point to extractable, non-commodity answers for typed questions. Multimodal generative AI impressions usually point to crawlable, representative visual assets on the same pages. Do not merge the two into one vanity total before you choose a fix.

This article gives you a Multimodal Response Ladder, a Search-Type Fit Scorecard, and a realistic worked example so you can decide which generative AI impression stream deserves the next sprint.

Key takeaways

• Google added web multimodal reporting to Search Console’s Search results and Generative AI performance reports on September 24, 2026. Multimodal covers image-led searches such as Google Lens, Circle to Search, image uploads, and Chrome’s “Search this image.”

• The Generative AI performance report (Search) still measures impressions in AI Overviews and AI Mode. It does not replace classic click analysis in the Search results Performance report.

• Multimodal traffic typically has no Queries tab because those searches mostly use images rather than typed text. Diagnose with pages, devices, countries, and paired Search results clicks.

• Default to text-based generative AI work when buyers research with typed commercial questions. Shift to multimodal asset work when image-led impressions concentrate on product, place, or visually distinctive pages.

• Impressions are not a promise of clicks, citations, or revenue. Use them to prioritize eligibility and source-page quality, not to claim AI ranking wins.

Why the new split changes the decision

For months, many teams treated “generative AI impressions” as one number. That was already incomplete because Search and Discover generative AI reports answer different questions. The multimodal search-type filter adds another split inside Search: typed web searches versus image-led web searches that can still surface your pages inside AI Overviews or AI Mode.

Google’s announcement is explicit about scope. Multimodal reporting covers searches where users search using images, including Lens, Circle to Search on Android, uploaded images, and Chrome image search. The same filter exists in the Generative AI features report, so you can see whether AI-feature impressions arrived from text or image-led journeys.

That matters because the corrective work is not identical. A thin commodity blog post can earn weak text-based AI visibility even if its hero image is beautiful. A product page with strong copy can still underperform in multimodal journeys if important visuals are CSS backgrounds, unlabeled, low resolution, or mismatched to the visible subject of the page.

What each Search Console view actually tells you

Keep the reports separate before you choose an action.

Two counting details prevent false alarms. In the Generative AI Search report, chart totals are aggregated by property, while the Pages table is aggregated by page. Google also notes that newest chart data can be preliminary. For AI Overviews and AI Mode, clicks on external links still count as clicks in Search Console’s broader methodology, but the dedicated generative AI report itself is impression-focused.

One more constraint shapes diagnosis: Google’s dimensions documentation states that specific text query data is not available for multimodal traffic because those searches mostly use images rather than text. If your team’s reflex is “pull the query report and rewrite titles,” that reflex only works cleanly on the text-based side.

The Multimodal Response Ladder

Climb these levels in order. Stop when the current level is unfinished. Do not jump to a media production project while eligibility is broken.

1. L0 — Confirm generative AI inclusion and report availability. In Search Console, confirm the property is included for Search generative AI features. Google’s generative AI report help notes that exclusion can explain a missing report or empty impression set. Also confirm the page is indexed and snippet-eligible; Google’s AI features documentation states those remain the technical baseline for supporting links in AI Overviews and AI Mode.

2. L1 — Split text-based and multimodal generative AI impressions. Open Performance → Generative AI. Export the last 28 days twice: once with Web: text-based and once with Web: multimodal. Record property-level impressions and the top 20 pages in each export. Do not average them into one “AI visibility” KPI.

3. L2 — Pair each stream with Search results clicks. For the same date range and search type, open the Search results Performance report. Note clicks and CTR for the overlapping URLs. A page with rising multimodal generative AI impressions and near-zero multimodal clicks needs a different review than a page with healthy text clicks and rising text-based AI impressions.

4. L3 — Score which stream deserves the next sprint. Use the Search-Type Fit Scorecard below. Pick one primary stream for 30 days unless both score high and share the same money pages.

5. L4 — Run one primary intervention queue. For text-based primacy: rewrite commodity answers, improve extractable structure, and strengthen unique evidence. For multimodal primacy: replace CSS-only visuals with crawlable img assets, add descriptive filenames and alt text, place images near relevant copy, and set a representative preferred image. Google’s image SEO guidance and generative AI optimization guide both support high-quality, relevant media as a supporting signal—not as a substitute for useful page content.

Search-Type Fit Scorecard

Score text-based and multimodal separately from 0–2 on each factor. Highest total becomes the primary 30-day queue.

Interpretation: 9–12 = primary stream for the next sprint. 5–8 = secondary monitoring. 0–4 = park it. If scores tie, choose the stream with stronger commercial page overlap and cleaner measurement. Progress you cannot observe usually becomes vanity work.

Worked example: a specialty retailer with two different “AI wins”

Hypothetical scenario (not an Oasbit client case study): a specialty outdoor-furniture retailer sees generative AI impressions rise for 28 days. Leadership assumes “GEO is working” and asks for more AI-optimized blog posts. A Search-type split changes the story.

• Text-based generative AI impressions concentrate on a buying-guide article and a materials comparison page.

• Multimodal generative AI impressions concentrate on three product detail pages with distinctive finishes.

• Search results Performance shows healthy text clicks on the comparison page, but weak multimodal clicks on two product pages whose main visuals are CSS background images.

Scorecard result in this hypothetical: text-based totals 9; multimodal totals 10 because money-page overlap and buyer visual discovery are stronger. The primary sprint is not another FAQ farm. It is making product images crawlable HTML images with descriptive filenames, useful alt text, nearby specification copy, and a preferred image that represents the product rather than a logo. The comparison page stays on a secondary queue for deeper non-commodity proof points.

That sequence follows Google’s own framing: generative AI features still rely on core Search systems, and supporting media helps when it is relevant—while commodity content and packaging tricks remain weak substitutes.

What to fix first on each winning stream

If text-based generative AI impressions win

1. Confirm snippet eligibility on the top impression URLs. Preview controls such as nosnippet can remove the page from the supporting-link path Google documents for AI Overviews and AI Mode.

2. Replace commodity summaries with unique process detail, comparisons, constraints, and first-hand evidence. Google’s generative AI optimization guide emphasizes non-commodity, people-first content over AI-only formatting tricks.

3. Keep important answers in crawlable text, not only in images or locked scripts.

4. Review classic text-based clicks and engaged sessions for the same URLs so impression growth is interpreted against visit quality.

If multimodal generative AI impressions win

1. Use real HTML image elements with a crawlable src. Google’s image SEO documentation states Google does not index CSS images.

2. Place high-quality, subject-accurate images near the relevant text on the landing page. Filenames and alt text should describe the subject without keyword stuffing.

3. Specify a preferred representative image with schema.org image or primaryImageOfPage guidance, or og:image, following Google’s preference for relevant, non-generic, non-extreme-aspect images.

4. Do not confuse the Images tab report with multimodal web. Multimodal is image-as-query inside web Search experiences, including generative AI features when those features show your links.

Common failure patterns

• One blended AI impression chart. Hides whether growth came from typed research or image-led discovery.

• Query-first diagnosis on multimodal rows. The Queries dimension is unavailable for that search type, so page and asset review must lead.

• Shipping decorative stock photos as a GEO project. Google’s guidance favors relevant supporting media, not unrelated decoration.

• Ignoring Search results clicks. Generative AI impressions alone cannot tell you whether people reached the site.

• Treating rising impressions as proof that a tactic “unlocked” AI rankings. Google warns that no third-party tool has access to internal ranking or AI systems, and eligibility is not a serving guarantee.

Limitations and exceptions

This framework is strongest when Google Search is a meaningful discovery channel and the Generative AI report is available for your property. Sites with too few generative AI impressions may not see the report yet. Multimodal metrics also appear only when the site receives qualifying image-led Search activity.

Some categories are visually led by nature—fashion, furniture, food, home services, local venues—while others remain text-led. A B2B software company can still see multimodal impressions, but buyer interviews may keep text-based work as the better primary investment.

Finally, multimodal generative AI impressions do not mean Google Lens “ranked” a photo the way a classic Images-tab result does. They mean your pages were shown as links inside generative AI features after an image-inclusive search. Interpret the metric as a visibility breadcrumb, then validate with clicks and on-site outcomes.

Recommended next steps

1. Export 28-day Generative AI impressions for Web: text-based and Web: multimodal separately.

2. Mark the top URLs in each export as commercial, informational, or brand support.

3. Complete the Search-Type Fit Scorecard and choose one primary stream for 30 days.

4. Ship five page-level fixes on that stream—answer quality for text, crawlable relevant visuals for multimodal—then re-check impressions and Search results clicks.

If you want help turning Search Console generative AI segments into a sequenced visibility plan across answer quality and visual source pages, review Oasbit’s generative engine optimization services. For reading the broader generative AI report before you change strategy, see how to read Search Console’s Generative AI report. When the split is clear and you need a practical 30-day roadmap, book a growth strategy session.

Sources

• Google Search Central Blog: Announcing web multimodal Search performance reporting in Search Console

• Search Console Help: Generative AI performance report (Search)

• Search Console Help: Performance report (Search results) overview

• Search Console Help: Performance report dimensions and data groupings

• Search Console Help: What are impressions, position, and clicks?

• Google Search Central: AI features and your website

• Google Search Central: Optimizing for generative AI features on Google Search

• Google Search Central: Google image SEO best practices