Google I/O 2026 + Gemini Omni: Archive Multimodal AI Outputs as PDF for AEO Workflows

TL;DR

Google I/O 2026 dropped two flagship models — Gemini Omni (any-input → any-output multimodal, including video) and Gemini 3.5 (frontier intelligence + action) — along with Antigravity 2.0, the all-new Antigravity CLI, WebMCP, and Google AI Mode upgrades. For anyone tracking how AI Overviews and AI Mode cite the web (an SEO problem now called AEO — Answer Engine Optimization), the announcements matter for one reason: the outputs from these models are increasingly multimodal (text + image + chart + video frame + table), and they need to be archived as portable artifacts. Convert: Web to PDF is the free Chrome extension that captures any Gemini, AI Mode, or ChatGPT output as a real PDF — selectable text, working links, locally-generated.

What dropped at Google I/O 2026

The keynote highlights, condensed:

Gemini Omni — multimodal model that takes any input (text, image, audio, video) and produces any output. Demoed editing video from a text prompt and synthesizing a full chart from a CSV in one shot.
Gemini 3.5 / Gemini 3.5 Flash — frontier intelligence with "agent" capabilities baked in. Flash variant runs faster than Gemini 3.1 Pro on coding and agentic benchmarks.
AI Mode — the conversational search experience inside Google Search, expanded to more queries and more multimodal answers
WebMCP — a proposed open web standard that exposes structured tools (JavaScript functions, HTML forms) to browser-based agents. This is the AI-browser plumbing.
Antigravity 2.0 + Antigravity CLI — agent orchestration platform with cross-platform terminal sandboxing
Voice features in Gmail, Docs, Keep
Google Pics — a new design tool
AI Ultra $100/month plan for developers and advanced creators
Native Kotlin in AI Studio for Android development
Updates to AI Inbox

For content creators, marketers, and AEO operators, the headlines that matter:

AI Mode is going to surface more of the web, and surface it as multimodal answers
The citations under those answers are the new "ranking" — which means tracking them is a core SEO task
Outputs need to be saved as portable, audit-able artifacts because they change session-to-session

Why archive Gemini / AI Mode outputs at all?

Four reasons:

1. Outputs are non-deterministic

Run the same prompt twice and you'll get two different answers — sometimes with different citations, different sources, different ordering. The "official answer" the AI gave you today isn't the same one it'll give you tomorrow. If you cited an AI output in a deliverable, you need a frozen artifact of what the AI actually said.

2. Citations are the new SERP

Inside Google AI Mode and AI Overviews, the citation chips beneath the synthesized answer are the closest analog to the old organic ranking. If your brand is being cited, that's the win. If it's not, you need to know what is being cited so you can compete. PDFs of AI Mode result pages — captured weekly — are the equivalent of a SERP rank-tracker for the AEO era.

3. Multimodal outputs don't paste cleanly

Gemini Omni's outputs include rendered charts, video keyframes, and structured tables alongside text. Copy-paste loses formatting. Screenshot loses text selectability. Real PDF preserves both visual and semantic layers in one file.

4. AI Ultra-tier sessions cost real money

If you're paying $100/month for the AI Ultra plan and running expensive multi-step agent workflows in Antigravity 2.0, the output of those workflows is an asset you've paid to produce. Treat it like one — archive it.

The AEO weekly capture workflow

Here's a workflow that takes 15-20 minutes a week and gives you a working AEO archive:

Step 1 — define your AEO target set

Pick 20-30 queries you want to be cited for. For a B2B SaaS: brand name + 10 product-category queries + 10 comparison queries + competitor brand mentions. For media: top 30 evergreen topics. For local: top 15 location-modifier queries.

Step 2 — run each query in Google AI Mode

Open google.com, ensure AI Mode is enabled (the toggle at the top of search results in eligible regions), run the query, scroll to see the full multimodal answer with citation chips at the bottom.

Step 3 — capture each AI Mode result as PDF

Click Convert: Web to PDF on the page. Article Mode usually strips too much (you want the chrome that shows the AI Mode context); leave it off, choose A3 paper size for long answers, capture.

The resulting PDF has:

The full AI Mode answer text (selectable)
The citation chips (clickable in the PDF)
The query and timestamp (in the header)
Any embedded multimodal cards (images, tables)

Step 4 — same for Gemini.google.com

For queries you ran directly in Gemini (not AI Mode), capture the conversation page. Gemini conversation URLs are stable per-conversation — the PDF + URL combination gives you a citable artifact.

Step 5 — same for ChatGPT (chat.openai.com)

ChatGPT search results, when you ran the same query via ChatGPT, are worth capturing in parallel — ChatGPT leads the AI search market share at 60.7% in 2026, more than Google AI Mode + Copilot combined. Capture both for full coverage.

Step 6 — file by week + query

Naming: aeo-archive/2026-W22/<query-slug>/<engine>.pdf. So:

aeo-archive/2026-W22/best-pdf-converter-extension/ai-mode.pdf
aeo-archive/2026-W22/best-pdf-converter-extension/gemini.pdf
aeo-archive/2026-W22/best-pdf-converter-extension/chatgpt.pdf

Step 7 — diff week-over-week

The week-over-week diff in citations is your AEO scoreboard. Did your brand get cited this week that wasn't cited last week? Did a competitor enter the citation pool? Did Google AI Mode swap its model under the hood and change what it surfaces?

Three reasons it matters specifically for AI Mode capture:

Capability	Chrome `Print → Save as PDF`	Convert: Web to PDF
Captures lazy-loaded multimodal cards	Often misses	Pre-scrolls to load all
Strips Google's interactive nav	No	Article Mode + element picker
Preserves citation chip URLs as clickable links	Sometimes	Yes
Handles the sticky "Ask follow-up" chat input	Awkward overlap	Removable with element picker
Lets you choose A3 / Tabloid for long AI answers	Yes	Yes
Captures the AI Mode model-attribution footer	Variable	Yes

For AI Mode pages specifically, the dynamic nature of the answer (the streaming text, the asynchronous citation chips, the lazy-loaded multimodal embeds) means a naïve print-to-PDF often misses content. The extension pre-scrolls and waits for content to load.

Gemini Omni multimodal outputs: PDF or video?

Gemini Omni produces video keyframes, animated charts, and audio summaries that can't fully exist as a PDF. For these outputs, two-track archiving:

PDF capture of the full chat / output page — preserves the prompt, all text responses, all clickable references, and visual frames of any video output
Source-file download — if Gemini produces a downloadable video or chart export, save that separately

Then if you need a single deliverable, the PDF references the source file by filename. The PDF lives in your repo or notion archive; the source file lives in a project folder. Don't try to embed the video in the PDF — file sizes balloon, and most PDF viewers don't render embedded video reliably.

What this means for content strategy

The AEO playbook in mid-2026:

Be the source the AI cites

Citations under AI Mode answers tend to over-index on:

Clean structured pages with H2/H3 hierarchy
Explicit definition-then-detail structure (the "answer-first" format)
Tables that summarize comparisons
FAQ schema that AI engines can extract as Q&A pairs
Date-of-update timestamps visible on the page

The "old" SEO playbook (backlinks, content velocity, keyword density) matters less than the "new" AEO playbook (content depth, scannable structure, citation-worthy claims with attributable sources).

Position 1 in classic SERP is becoming less valuable because the AI Overview eats the click. Citation chip presence is more valuable. Track:

% of target queries where your brand appears in a citation chip
Share-of-citation in your category (you vs top 3 competitors)
Week-over-week change

The captured PDFs are the raw data. Process them however your team likes — manually, with an internal LLM, or with a structured scrape.

Be specific in claims

AI Mode strongly prefers content with concrete, attributable claims — "X processes 100MB files in 4 seconds" beats "X is fast." Specific numeric claims, version numbers, dates, and named methodologies are what gets extracted. If you've been hedging in your copy, the AEO era is the time to stop.

A 30-minute audit you can do today

If you haven't started AEO tracking yet:

Pick your 10 highest-priority queries
Run each in Google AI Mode, capture as PDF
Run each in ChatGPT search, capture as PDF
Run each in Perplexity, capture as PDF
Tag the PDFs by: did your brand appear? which competitors appeared? what kind of source got cited (publisher, doc, forum, repo)?
Add to your weekly cadence

Total time: 30-45 minutes. You'll have a baseline by end of day.

Antigravity CLI agent runs: a special case

The new Antigravity CLI from Google I/O lets you spin up subagents in cross-platform terminal sandboxes. Long-running agent workflows generate logs — the chain of tool calls, intermediate reasoning, file diffs, and final outputs. Those logs are also a deliverable. We have a separate post on archiving Antigravity logs as PDFs with Convert: Anything to PDF (drop the Markdown log file in, click convert). For the web outputs — the dashboard view of an Antigravity run — the Web-to-PDF extension is what you want.

Why local archiving (not "cloud notebook")

Tempting answer: "Just paste everything into a Notion or Google Doc." Two problems:

1. Source attribution gets stripped

When you paste an AI Mode page into Google Docs, the citation chip URLs sometimes survive, sometimes don't, depending on copy-source. The structural hierarchy gets flattened. The PDF preserves both.

2. Cloud notebooks are tied to platform

If you change company, change tools, or lose Notion access, the archive evaporates. Local PDFs in a versioned aeo-archive/ directory (in your team's Git repo, S3 bucket, or shared drive) don't depend on any vendor's continued service.

A note on AI model comparison

For comparing how different AI engines answer the same query — useful both for AEO research and for personal AI literacy — pair the captured PDFs with a comparison table. The pattern that works:

One PDF per (query, engine) pair
A spreadsheet that lists query × engine × cited sources × citation type
A short narrative summary per query in a separate doc

For tracking taste in AI-recommended content (different problem — streaming, not search), CineMan AI overlays IMDb / Rotten Tomatoes / personal taste-match scores directly on Netflix and Prime Video. Same studio, same indie shop posture, free.

Frequently asked questions

Q: How do I get access to Gemini Omni or Gemini 3.5?

Both rolled out in tiers through I/O week. Free tier of Gemini.google.com gives base access; Gemini Advanced ($20/mo) and AI Ultra ($100/mo) unlock heavier multimodal usage and Antigravity tier features. Check gemini.google.com for current eligibility.

Q: Will Google's AI Mode citation chips work as clickable links inside the captured PDF?

Yes — Convert: Web to PDF generates real PDFs (not screenshots), so the underlying <a href> elements are preserved as clickable hyperlinks. Open the PDF in any modern viewer and the citation links work.

Q: Can I capture a Gemini conversation that's hundreds of turns long?

Yes. The extension pre-scrolls to load the full conversation history and captures the whole thing. For very long conversations, use a larger paper size (A3 or Tabloid) so the output fits cleanly without micro-fonts.

Q: Does this work for ChatGPT and Claude.ai too?

Yes — both render their conversations as standard web pages. Capture works the same way. For Claude.ai specifically, the side panel and artifact rendering both come through.

Q: How is this different from just exporting from the AI's own interface?

Gemini and ChatGPT have "export" options that often produce stripped-down text or HTML files without the citation chips or multimodal embeds. The browser-rendered version (which you see) includes everything; capturing that as PDF gets you the complete artifact, including model-attribution footer and timestamp.

Q: Are AI outputs sensitive enough that I shouldn't use an online URL-to-PDF tool?

Often yes. If your prompt contained sensitive business context (and at $100/mo AI Ultra workflows, it almost always does), don't route the output through a third-party "convert URL to PDF" service. Locally-running Chrome extensions like Convert: Web to PDF make zero network requests during conversion.

Q: What's the file size of a typical AI Mode capture?

200 KB – 1.5 MB depending on multimodal content density. Pure-text captures stay under 300 KB. Captures with embedded cards (charts, image grids) can grow to 2 MB.

Q: Can I batch this across 50 queries at once?

The extension is one-page-at-a-time; for batch automation you'd want a Playwright/Puppeteer script. The manual workflow is fast enough for weekly cadence: 50 queries × 30 seconds = 25 minutes.

Q: What's WebMCP and does it change archiving?

WebMCP is a proposed open standard for exposing in-page tools to browser-based agents. It changes how agents act on a page but doesn't fundamentally change archiving — the page is still HTML, still capturable as PDF. Worth watching as it matures.

Q: Will Gemini outputs include EXIF metadata I should be careful about?

Possibly — Gemini Omni's generated images can include synthesis metadata (model version, prompt hash). If you're publishing the PDF externally, decide whether that metadata stays embedded. The extension preserves what's on the page; for stripping image metadata, post-process separately.

Q: What about competitors like Perplexity Pro and Comet?

Same workflow. Capture each engine's answer for the same query, file alongside the Google / Gemini / ChatGPT captures. The diff across engines is itself a useful AEO signal — engines that agree on a citation are likely to agree about the underlying source's authority.

It shifts the calculus toward depth and citation-worthiness. If you have 1000 thin pages that ranked well in 2023, expect them to lose visibility through 2026. The pages that get cited in AI Mode are the longer, more specific, more structured ones — same hands-on detail that gets a developer to share a post on Hacker News.

Bottom line

Google I/O 2026 made it clearer than ever that AI Mode and Gemini Omni are the new front page of the web, and citation chips are the new ranking. Track them by capturing each weekly query result as a real PDF with Convert: Web to PDF — selectable text, working links, locally-generated, free. Build the archive now; the week-over-week diff is what teaches you what AEO actually rewards.