AI Search Is Still Search. Seeing It Is the Hard Part.

· Logan B. Jones-Olson

We checked 13 popular AI search tactics against platform docs and published studies. Three hold up, three have no support, and measurement is the bigger gap.

Abstract illustration in slate and gold: faint search paths across a grid, with a few gold lines reaching ringed nodes and most paths fading out before they arrive

The short answer

As of October 30, 2026, the evidence points one way. AI answer engines pull from search indexes and reward what search has rewarded for years: pages a crawler can read, content nobody else has, and a reputation that exists beyond your own site. Several AI-specific tricks, like llms.txt files and schema added to earn citations, have no measured effect on whether you get cited. Most of the rest help only as far as they make a page easier for people to read.

What changed is harder to see. Fewer people click. Many of the ones who do arrive from an AI answer show up with no referrer, so analytics files them under “direct.” If you judge AI search by the AI referral line in your dashboard, you are looking at a floor, not the total.

What the platforms actually say

Google put its position in writing. Its guide to generative AI features in Search says “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” The same guide says an llms.txt file “will neither harm nor help” a site’s visibility in Google Search, that “there’s no requirement to break your content into tiny pieces,” and that “there’s no special schema.org markup you need to add.”

Google also draws a line between commodity and non-commodity content. Its example of commodity content is “7 Tips for First-Time Homebuyers.” Its example of the other kind is “Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line,” a first-hand account nobody else could write.

Microsoft is more prescriptive. Its October 2025 guidance on getting content into AI search answers recommends clear headings, lists, tables, Q&A formats and schema, and warns that AI systems may not render content hidden in tabs or expandable menus. The two companies disagree on structure and schema. They agree that a crawler has to be able to read the page.

OpenAI and Anthropic each run separate crawlers for model training, search, and fetches a user asks for. OpenAI’s documentation says sites that block its OAI-SearchBot “will not be shown in ChatGPT search answers.” That robots.txt decision has a clear, documented cost.

13 tactics, checked against the evidence

These tactics show up again and again in “AI-first website” checklists this year.

Scorecard of 13 AI search tactics. Holds: avoid client-side rendering; publish original data, sources and dates; separate training, search and user bots in robots.txt. Partly holds: serve Markdown to AI bots; chunk content into 256 to 512 tokens; answer in 30 to 60 words after every heading; make every section stand alone; use tables and lists over prose; write headings as questions; build agent endpoints. Unsupported: publish llms.txt; flatten URLs to three clicks; add nested entity schema for citations.
Three hold up, seven partly hold, three have no support. Details and sources below.
# Tactic Verdict What the evidence shows
1 Publish llms.txt and llms-full.txt Unsupported Google says the file neither helps nor harms. Google’s John Mueller wrote in June 2025 that, at the time, “no AI system currently uses llms.txt.” Ahrefs found 97% of about 38,000 published llms.txt files got zero traffic in May 2026.
2 Serve Markdown versions to AI bots Partly holds Cloudflare converts pages to Markdown when a request asks for it, and says Claude Code and OpenCode already ask. Those are coding agents, not the search crawlers behind citations. Showing bots different content than people see is a cloaking risk.
3 Avoid client-side rendering Holds Vercel’s server-log research found “none of the major AI crawlers currently render JavaScript.” Googlebot and Applebot do. If your key content needs JavaScript to appear, most AI crawlers never see it.
4 Keep every page within three clicks Unsupported No platform or dataset ties click depth to AI citations. Crawlers find URLs through indexes, sitemaps and links. Ordinary internal linking still applies.
5 Chunk content into 256 to 512 tokens Partly holds Engines retrieve passages, not whole pages. No consumer AI search product publishes a chunk size, and Google says chunking isn’t required.
6 Answer in 30 to 60 words after every heading Partly holds Kevin Indig’s analysis of 18,012 ChatGPT citations found 44.2% came from the first 30% of a page. That is a correlation, for one engine. The 30 to 60 word rule has no primary source. Leading with the answer is good writing anyway.
7 Make every section stand on its own Partly holds Fits Microsoft’s guidance and passage-level retrieval. No controlled study tests it.
8 Use tables and lists over prose Partly holds Microsoft recommends them. Google says to organize pages for readers. Breaking everything into fragments hurts the people reading.
9 Write headings as questions that match prompts Partly holds Clear headings help. Google says it understands synonyms, so mirroring every prompt variant is wasted work.
10 Add nested entity schema to earn citations Unsupported In Ahrefs’ controlled test, 1,885 pages added schema against 4,000 matched controls. Citations moved 2.4% in AI Mode and 2.2% in ChatGPT, neither meaningfully different from zero, and fell 4.6% in AI Overviews. Basic Organization and Article schema still earn rich results.
11 Publish original data, sources and dates Holds Google prioritizes first-hand, non-commodity content. Ahrefs found AI assistants cite content 25.7% fresher than Google’s organic results. In a lab benchmark, the GEO paper (Aggarwal et al., KDD 2024) found that adding citations, quotations and statistics raised visibility by up to 40%.
12 Separate training, search and user bots in robots.txt Holds OpenAI and Anthropic document a separate crawler for each job. Blocking the search crawler removes you from that engine’s answers. Allowing training is a business decision.
13 Build agent endpoints for booking and pricing Partly holds The standards for agents to book and buy are still being built. Worth a pilot for ecommerce and booking-heavy businesses. Early for most services firms.

Where citations come from

Answer engines lean on a short list of third-party sites. Profound’s study of 680 million citations, from August 2024 to June 2025, found Wikipedia made up 7.8% of all ChatGPT citations. Reddit made up 6.6% of Perplexity’s citations and 2.2% of Google AI Overviews’.

That mix moves fast. Reddit averaged 3.83% of ChatGPT Search citations from July 18 to August 7, 2026, fell below 1% by August 14, and averaged 0.52% after that, per Promptwatch data reported by Axios. Promptwatch called the size of the drop provisional. A plan built around one engine’s favorite source this quarter can break next quarter.

The part that holds across engines: you need to exist in places other than your own site. For a B2B firm, my starting list is named people at your company writing on LinkedIn, podcast appearances, contributed articles, directory and review profiles, and a spot in the lists buyers already read.

The click gap

Fewer searches turn into visits.

Cited brands got slightly more clicks. In Seer’s data, brands cited inside the AI Overview had a 0.70% organic click-through rate, vs 0.52% when they were not cited. Seer notes that this does not prove the citation caused the difference.

Diagram of where AI search influence shows up in analytics. Someone reads an AI answer that mentions you, and four things can happen: they click a cited link and the referrer comes through, which shows up as an AI referral; they click from a native app that sends no referrer, which shows up as direct traffic; they search for you days later, which shows up as organic search; or they never click, which shows up only in Search Console and Bing Webmaster Tools reports.
AI search influence often shows up in channels that don't say "AI." Sources: Cloudflare, Similarweb, Google, Microsoft.

Why your analytics miss most of it

This is the part I care about most, because I have fought it before. When I joined Linqto, more than 80% of conversions reported as “direct.” At that scale, “direct” means stripped referrers, untagged links, dark social, and redirects dropping data on the way in. Rebuilding attribution cut unattributed “direct” traffic to under 30%. The full case study covers how.

AI search feeds that same bucket in three ways.

  1. Visits with no referrer. Cloudflare found that traffic from Claude’s native app carries no Referer header, and believes the same holds for other native apps. Similarweb confirms that visits without a referrer go into direct traffic.
  2. Visits that come later, through search. Similarweb found people who got a brand recommendation from an AI assistant were 2.5 times more likely to visit that brand’s site within seven days. Search accounted for about 56% of those visits, vs about 40% of typical visits. Someone who reads about you in an AI answer on Tuesday and searches your name on Friday shows up as organic search.
  3. Impressions with no click. Google’s Search Console report for generative AI features shows impressions in AI Overviews and AI Mode, with no clicks. Bing Webmaster Tools’ AI Performance report shows citations and a sample of grounding queries, also without clicks.

So treat the AI referral line in your dashboard as a floor. Similarweb counted an average of 770.7 million AI referral visits a month worldwide from June 2025 to May 2026, up 117.4% from the year before. Visits without a referrer fall outside that count, so it can only miss AI traffic, never add to it.

What to measure instead

What we did on this site

lbjoconsulting.com is a static site: every page is plain HTML on the first request, with nothing important behind JavaScript. Our robots.txt allows search, user and training crawlers on purpose. We verified the site in Google Search Console and Bing Webmaster Tools on October 18, 2026, submitted the sitemap to both, and sent every URL to IndexNow. We keep an llms.txt file as a cheap hedge, knowing Google says it neither helps nor harms in Search.

That setup is twelve days old as I write this, so there is no data worth sharing yet. We will update this piece in late January 2027 with what Search Console, Bing and our own referrer data show after 90 days.

What I would do this quarter

Do now

Cheap hedges

Skip

If you want this applied to your site, our AI search visibility audit covers crawler access, measurement, and the content and off-site gaps, in that order.

Changelog

Sources

Want this thinking applied to your business? Schedule a free consultation or see the services.