How Can You Stop Traffic Loss to AI Search?

AI search visibility: stop chasing outputs, fix your inputs

Zeno Rocha’s “Control how you show up in AI search” is the right nudge at the right time: you don’t “optimize” the answer box, you make your site the clean, citeable source it wants to use. Here’s how I’d run that play with an input-control checklist and a 30‑day test loop. 🎯

// My reaction

Excited nod

If your traffic’s drifting to AI answers, this is worth talking about. Google keeps expanding AI Overviews/AI Mode (May 14, 2024 rollout; AI Mode introduced Mar 5, 2025), says usage is up (“more than a billion people” using AI Overviews; “people are using Google Search more than ever,” Mar 5, 2025), and claims links in summaries help publishers via “higher-quality clicks” (May 30, 2024; also see “driving more queries and higher-quality clicks,” 2025). Independent studies show the opposite on clicks (Pew, Jul 22, 2025; field/behavioral papers 2026, another 2026 experiment). Both can be true by query class. Your move is the same either way: be the shortest path to correct facts, in code and in copy.

// My reaction

Concerned surprise

What this is really about

You can’t “optimize” AI answers directly. You control inputs. That means: entity clarity, extractable formats, explicit claims with dates, and clean citations. If your site isn’t a source of truth with machine-readable context, AIs will quote someone else who is. 🧩

If you want the deeper mechanics of why LLMs pick sources the way they do, skim my write-up on retrieval signals and content structure: AI Search Information Retrieval Primer. Short version: structure and citations beat prose, every time.

The five moves I actually run

1) Entity hygiene and canonical pages

• One canonical “About” page with legal name, alt names, NAP, founders, founding year, and persistent IDs like Wikidata if you have it. Add sameAs links to LinkedIn, Crunchbase, GitHub, Google Business Profile, and the rest. Keep it boring and factual.
• Service and location canonicals: one page per service, one per location or service area with boundaries, hours, and contact. Don’t bury this in hero videos.
• Why it matters: models resolve entities before quoting. Messy or split identity means lost attribution.

2) Make content extractable, write for token parsers and humans

• Answer first: lead with a one-sentence definition or outcome, then details. See: Build Content for AI Search.
• Use lists, tables, FAQs: models latch onto structure. Include constraints, ranges, and examples.
• Timestamp claims: expose dates in HTML and JSON‑LD using datePublished and dateModified. This helps both selection and reconciliation.

3) Schema that carries your claims, not fluff

• Organization or LocalBusiness: name, alternateName, sameAs, foundingDate, address, geo, contactPoint.
• Service or Product: serviceType or category, areaServed, offers with real price or priceSpecification if you can publish ranges.
• FAQPage or HowTo: only when the page truly is a FAQ or a procedure. Google has pulled back rich results in the past, so keep these for clarity, not tricks.
• Validate your JSON‑LD. If you need a quick gap scan, try the Schema Opportunity Analyzer.

4) Crawlability and freshness signals

• Sitemaps: complete, current, with lastmod. Submit in GSC. Keep orphan URLs out.
• Robots.txt: allow public pages, block junk. Many AI crawlers honor robots directives, but do not treat that as privacy, treat it as clarity.
• Stable URLs: avoid frequent changes. You’ll break embeddings and lose historical context that supports attribution.

5) Prove you’re quotable

• Citations and sources: link to your primary data, PDFs, or research pages. If you use external stats, cite them cleanly. 📈
• Policy pages: licensing, terms, editorial standards. AIs prefer sources that look accountable, and users do too.
• Monitor AI answers: run a monthly check on priority prompts. My quick start is the AI Visibility Checker. Save the queries that map to your sales funnel and watch who gets cited.

A compact checklist you can run this week

How I’d test whether this worked

• Define five buyer-intent prompts and five authority prompts for your niche. Example: “best fractional CFO pricing for SaaS,” “how to set up accrual accounting for seed-stage startups,” and so on.
• Capture baseline AI answers across Gemini, ChatGPT, and Perplexity. Note who’s cited and what language they quote.
• Ship the five moves above for your core pages.
• Recheck in 30 days. You’re looking for your brand cited, your phrasing echoed, or your page in the Sources. If nothing moves, your claims aren’t concrete enough or another entity is cleaner.

What not to do

• Don’t stuff FAQ schema sitewide. AIs will still paraphrase, humans will stop trusting you.
• Don’t hide critical facts in images or PDFs only. Put the claim in HTML, then link the PDF as a source.
• Don’t publish city-page templates with swapped nouns. That’s noise, not clarity.

Takeaway

You influence AI answers by being the cleanest, most explicit source, not by gaming prompts. Get your entity straight, write extractable claims, mark them up, and keep score. Same muscle that wins classic SEO, just a different interface reading first. Build pages for machines and people without fluff, then measure wins where it matters in SEO, GEO, and AI practice.