Quick Takeaways (What to change this week)

• Stop thinking “page optimization.” Start designing 90–140 word citation paragraphs that can stand alone (median highlight ~117 words).
• Use an answer‑first pattern: sentence 1 = direct answer; sentences 2–5 = constraints, steps, proof.
• Add 1–3 “citation blocks” per money page (definitions, comparisons, rules/eligibility, step‑by‑step).
• Don’t measure AI traffic by #:~:text alone. Reports suggest this signal dropped after May 2026 as citations shifted to inline links—so you need a multi‑method attribution plan.
• Operationalize a weekly loop: target pages → insert/upgrade citation blocks → re‑test with a consistent query set → log which passage gets cited.

What “scroll‑to‑text” citations actually are (and why they change AEO)

A text fragment link is a URL that includes a special fragment identifier like #:~:text=. When clicked in supported browsers, it scrolls to the matching text and highlights it—often in a distinct color (many people describe it as a purple highlight).

From a marketer’s perspective, this is not just a UX detail. It’s a citation unit. If Google AI Mode is deep‑linking to a highlighted passage, it’s effectively saying: “This exact paragraph is the evidence for my answer.”

If you want the technical background, Google’s Chrome team documented how these links work in Text Fragments on web.dev—including how the browser finds and highlights the target string. The key implication: your paragraph’s wording and structure becomes the “address” Google can reliably point to.

Why this is a passage-level shift (not a new acronym)

We’ve spent years optimizing pages for ranking and snippets. But AI Mode citations are behaving like “passage retrieval.” The Search Engine Land analysis of 15.7 million AI Mode citations found:

• Nearly half of citations were scroll‑to‑text deep links.
• The median highlighted span was ~117 words (think: a full paragraph).
• ~80% of winning passages answered in the first sentence.

This is why “highlight‑first AEO” is so actionable: you can deliberately create paragraphs that are easy to extract, hard to misquote, and clearly attributable.

The “Highlight‑First” paragraph template (90–140 words)

Here’s the on‑page pattern we recommend when you want Google AI Mode (and other answer engines) to cite your passage. It’s designed to fit the observed ~117‑word median highlight and the answer‑first tendency.

Template: Answer → Boundaries → Steps/Criteria → Proof

1. Sentence 1 (Answer): Direct, unambiguous, query‑matching statement.
2. Sentence 2 (Boundaries): When it applies / doesn’t apply (time, region, product tier, prerequisites).
3. Sentences 3–4 (Steps or Criteria): 2–4 concrete steps, thresholds, or decision rules.
4. Sentence 5 (Proof): Evidence: metric, source, tool output, or a verifiable reference.

What this looks like in practice (Example #1: definition block)

Good “citation‑ready” paragraph (≈115 words):

Scroll‑to‑text citations are deep links that highlight a specific passage on a page using #:~:text=. Google AI Mode uses them to point readers to the exact paragraph that supports its answer, not just the top of the page. They work best when the highlighted passage is self‑contained: define the term, state the key implication, and include one concrete detail (like a typical word range or a step). In our testing, the most consistently cited passages start with the definition in the first sentence, then add 2–4 sentences of context and constraints so the paragraph still makes sense when read in isolation.

Bad paragraph (hard to cite): long preamble, pronouns (“this,” “it”), references to earlier sections (“as mentioned above”), and missing boundaries. AI systems avoid lifting text that becomes confusing out of context.

Micro-rules that increase citation “liftability”

• Use the query’s nouns in sentence one. If the query is “AI Mode scroll‑to‑text citations,” put that phrase (or close variant) in the first sentence.
• Avoid vague openers: “It depends,” “There are many ways,” “In today’s world.”
• Keep it self-contained: no “see below,” no table references, no “this guide.”
• Include at least one number when relevant (thresholds, time, size, steps, limits).
• Prefer active voice and explicit subjects: “Google AI Mode cites…” not “Citations are made…”

Where to place “citation blocks” (so Google has something to grab)

The biggest mistake we see: teams add one great paragraph… buried halfway down a page that starts with a 600‑word story. AI Mode can still find it, but you’re making extraction harder than it needs to be.

Instead, add 1–3 dedicated citation blocks per high‑value page in predictable locations. This aligns with the idea from the 15.7M citation study that you can intentionally create extractable units (Search Engine Land).

High-performing block types (pick based on intent)

• Definition block: “What is X?”
• Comparison block: “X vs Y—what’s the difference?”
• Rules/eligibility/pricing block: “You qualify if…” / “Pricing includes…”
• Step-by-step block: “How to do X in 4 steps.”
• Troubleshooting block: “If X happens, do Y.”

Example #2: comparison block (≈125 words)

AEO is about getting your content quoted as the answer; SEO is about ranking the page that contains the answer. In Google AI Mode, that difference matters because citations often point to a specific paragraph (~117 words) rather than the full page. Use SEO to win crawlability, indexation, and authority signals; use AEO to format the exact passage an AI system can reuse without losing meaning. Practically, that means adding an answer-first paragraph near the top, using explicit criteria (e.g., “use 90–140 words, include 2–4 steps”), and backing claims with verifiable sources (Search Console data, pricing pages, documentation). If your paragraph can stand alone in a screenshot, it’s usually citeable.

Example #3: rules/eligibility block (great for SaaS and marketplaces)

To track Google AI Mode-driven sessions in GA4, you can’t rely only on #:~:text appearing in landing page URLs. Community reports suggest Google changed AI Overview link formats around May 2026, reducing the consistency of text fragments as an attribution signal. Instead, treat AI traffic as a blended segment: monitor landing pages that are frequently cited, compare engagement deltas after citation-block updates, and validate visibility through controlled prompt checks in AI Mode. This approach catches both “highlight” deep links and newer inline citation links, so your reporting doesn’t break when Google changes URL formatting.

Measurement fix: why #:~:text is now unreliable (and what to do instead)

For a while, some teams used text fragments as a convenient filter: “If the landing page contains #:~:text, it must be AI Overview/AI Mode traffic.” That was never perfect—but it was directionally useful.

Then the ground shifted. A widely discussed thread reports that after a May 6 update, #:~:text was no longer sent from AI Overview links for some sites, likely due to citations moving to inline links or different link architecture. See: the Reddit discussion on tracking AI Overview traffic after the May update.

The new rule: optimize for highlight-first, measure with redundancy

Text fragments are still an optimization target when present (because they reveal the exact cited passage). But they’re no longer a dependable analytics identifier. So your measurement plan needs at least three overlapping signals:

1. Prompt-based citation sampling (visibility validation)
2. Landing-page engagement tracking (behavior validation)
3. Traffic-source dimension analysis in GA4 (attribution triangulation)

Step-by-step: a practical attribution workflow you can run weekly

Step 1: Build a controlled query set (20–50 prompts)

Create a stable list of queries you want to win. Keep it consistent week-to-week so you can detect drift. Include:

• 10 head terms (category-level)
• 10 mid-tail comparisons (“X vs Y for Z”)
• 10 “how-to” tasks
• Optional: 10 support/troubleshooting queries (these often trigger citations)

Step 2: Record citations at the passage level (not just “did we appear?”)

When AI Mode cites you, capture:

• Query
• Landing page URL
• Whether it used #:~:text (yes/no)
• The exact highlighted text (copy/paste the cited paragraph)
• Position/context (is it the primary answer or a supporting citation?)

This is where the “highlight-first” mechanic is gold: it tells you which paragraph is winning. If Google stops sending the fragment, you still keep the “exact quoted passage” record.

Step 3: Tie visibility to behavior (GA4 engagement deltas)

In GA4, create an exploration that compares engagement on “target landing pages” before vs after you add citation blocks. Focus on:

• Engagement rate
• Average engagement time
• Key event rate (trial start, demo request, add-to-cart)
• Scroll depth (if you track it via GTM)

Step 4: Use traffic-source dimensions correctly (avoid misreads)

When you analyze traffic sources, make sure you understand what GA4’s dimensions actually represent (and how they change across scopes). Google’s documentation on traffic-source dimensions is worth bookmarking—especially when you’re trying to separate “google / organic” from referrals, cross-network, or unassigned traffic that may appear when link formats change.

Step 5: Keep a second engine as a canary (reuse and drift detection)

AI Mode isn’t the only system doing passage retrieval. Run the same query set in at least one other answer engine (e.g., Perplexity) to see whether your paragraph structure is generally “liftable.” If your citation blocks get reused elsewhere, you’ve likely hit a robust pattern—not a Google-only quirk.

If you want a tighter cadence, we recommend pairing this with our internal workflow from GSC’s Generative AI Report: 30‑Min Citation Audit Loop, then layering passage-level logging on top.

Common mistakes (and how to fix them fast)

Mistake 1: Writing “teaser” paragraphs that require context

Symptom: Your paragraph starts with “This means…” or “In this section…” and gets ignored.

Fix: Rewrite sentence one so it can be quoted alone. Replace pronouns with explicit nouns (product name, feature name, rule name).

Mistake 2: Over-optimizing for short snippets

Some teams still compress answers into 1–2 sentences. But if the median cited span is ~117 words, you often need:

• the answer,
• the condition(s), and
• the “how” or proof

Fix: Aim for 90–140 words so the paragraph contains enough context to be trustworthy.

Mistake 3: Using #:~:text as your only measurement signal

Symptom: Your “AI traffic” report collapses to near-zero after a link architecture change.

Fix: Adopt the redundant measurement stack described above, and treat text fragments as a diagnostic clue—not your tracking foundation. The May 2026 change discussed by practitioners is a perfect example of why (thread here).

Mistake 4: Not separating “visibility” from “traffic” in reporting

AI citations can increase visibility without proportional clicks (especially when the answer is fully satisfied in the SERP). If you’re seeing confusing trends in Search Console totals, you’ll want a two-lens model.

Fix: Pair this post with Google Quietly Changed GSC Totals: AI Mode Counts Now so you don’t misattribute impression shifts to “SEO performance” alone.

Implementation: upgrade an existing page in 30 minutes

Here’s a fast, repeatable way to retrofit citation-ready paragraphs onto pages that already rank (or already have authority) but aren’t getting cited.

Checklist (do this in order)

1. Pick one target query you want the page cited for (not 10).
2. Identify the extraction point: where would an AI system grab the answer? (If you can’t point to a paragraph, neither can it.)
3. Write one 90–140 word answer-first paragraph using the template.
4. Add one proof element: a number, a tool output (GSC metric), or a reference to official documentation.
5. Place it high (top third of the page) unless the intent is troubleshooting (then place near the relevant section).
6. Re-test with your controlled query set and log citations.

What “proof” can look like (without turning the paragraph into a bibliography)

• Documented behavior: cite a platform doc (e.g., GA4 traffic-source definitions).
• Observed metric: “Median highlight span ~117 words” (from the 15.7M analysis).
• Tool-based verification: “In Google Search Console, this page gained X impressions after update.”

One more practical tip we’ve found: if you have multiple intents on one page (definition + pricing + setup), you’ll do better with multiple citation blocks rather than one mega paragraph that tries to answer everything.

FAQ: passage-level citations and tracking

How long should a “citation-ready” paragraph be?

Aim for 90–140 words. The large-scale analysis reported a median highlighted span of ~117 words, which maps cleanly to a well-structured paragraph with answer + constraints + steps + proof.

Do I need to force #:~:text links on my site?

No. Google generates these when it chooses to deep-link to a passage. Your job is to make sure there’s a clear, self-contained paragraph worth deep-linking to.

Why did my text fragment traffic disappear?

It may not be “traffic disappearing”—it may be the URL format changing. Practitioners have reported that after May 2026, AI Overview links sometimes stopped passing #:~:text, likely due to inline link changes (discussion). That’s why your measurement needs redundancy.

Does this replace SEO best practices?

No—passage-level AEO sits on top of traditional SEO. If crawling/indexing fails, your perfect paragraph won’t get cited. If you suspect infrastructure issues, our guide on crawl blocks is a useful companion: Your robots.txt Allows AI—But Your CDN Blocks Crawlers.

What to Do Next (Action Steps)

1. Choose 20 “citation target” pages (product pages, pricing, high-intent guides).
2. Add 1–3 citation blocks per page using the 90–140 word answer-first template.
3. Run a weekly prompt-based citation audit and store the exact cited paragraph text.
4. Update GA4 reporting to avoid relying on #:~:text as a sole identifier; use engagement deltas + traffic-source dimensions.
5. Re-test and iterate—treat each cited paragraph as a “unit” you can refine like ad copy.

If you want to systematize this across your site, try our AEO tool dashboard—sign up at https://aeotool.ai/register. We also recommend installing our Chrome extension for quick on-page checks and audits: AEO Analyzer Chrome extension.