Free report · July 2026

What actually works in AI search in 2026 and what’s expensive hype.

ChatGPT, Perplexity and Google’s AI answers are changing how customers find businesses like yours. Most advice about it is guesswork, and some of it is simply made up. This free report sets out what the published data actually shows: what is worth doing, what is not, and how to tell the difference.

Free PDF · instant download · no spam, unsubscribe in one click.

Read it in an afternoon. Save yourself from paying for things that do not work.

Free reportW
The 2026 AI Search Report
What works, what doesn’t, and what’s hype.
Inside
01The ten headline findings
04What actually works, graded by evidence
05The hype list: what to ignore in 2026
06Quick wins you can run this week
65 pages · 13 chapters July 2026
Free download
The 60-second version

How AI search actually works now

There is not one “AI search.” There are four surfaces your customers now use, each built differently, each citing a mostly different set of pages.

Google AI Overviews

The AI summary box that now sits above the blue links, built from Google’s normal index. It hits your traffic hardest because it answers the question before anyone scrolls.

7.3% → 1.6%
#1 organic click-through rate when an AI Overview appears
Ahrefs, 300,000 keywords

Google AI Mode

The chat-style version of Google. Made the default search experience worldwide in May 2026 — impressive on paper, but still a small slice of real behavior.

0.34%
of US searches, despite being the new default — about one in 300
SparkToro / Similarweb

ChatGPT

The standalone assistant most people mean by “AI.” The giant of AI referrals, leaning heavily on Bing’s index and weighting your opening paragraphs.

900M
weekly users, sending about 80% of all AI referral traffic
OpenAI-reported / referral data

Perplexity

A smaller answer engine with its own index and a loyal, high-intent audience. Worth staying visible in — it costs nothing — but not worth a big budget line.

2.0% → 1.3%
share of global web traffic, March to May 2026 — small and shrinking
Similarweb

One thing every surface shares: it can only recommend a page it can find, fetch, and read. That is where the real work is — and it is cheaper than what you are being sold.

The headline numbers

Ten findings from the report

Each of these comes from a named, published study. The report gives the full context behind every one.

68%
of Google searches now end without a click — up from ~49% in 2019. When an AI summary appears, people click through 8% of the time versus 15% without one.
SparkToro / Similarweb · Pew Research, 68,879 searches
7.3% → 1.6%
Number-one-position click-through rate on queries where an AI Overview triggers.
Ahrefs, 300,000 keywords
0.1–1.1%
AI chatbot traffic is still a rounding error of most sites’ traffic. Google organic out-sends every AI engine by 47x to 200x.
Search Signal / Ahrefs / Similarweb
Zero
Schema markup moved AI citations by zero. A controlled test of 1,885 pages found AI Overviews −4.6%, AI Mode +2.4%, ChatGPT +2.2% — statistically indistinguishable from nothing.
Ahrefs difference-in-differences study
97%
llms.txt is ignored. Across 137,000 domains, 97% of llms.txt files got zero AI-crawler requests. Google says outright it does not use them.
Ahrefs · Google Search Central
~3x
Brand mentions beat backlinks. Off-site mentions correlate with AI Overview visibility at r=0.664 versus r=0.218 for backlinks.
Ahrefs, 75,000 brands
44.2%
of AI citations come from the first third of a page. The single most valuable paragraph you own is the first one — and it is free to fix.
Evertune, ~400M citations
Under 1%
Ask the same question twice, get the same brand list under 1% of the time — the same order under 0.1%. Nobody can “guarantee” an AI ranking that will not hold still.
SparkToro / Gumshoe.ai, 2,961 runs
51%
of B2B software buyers now start their research in an AI chatbot — up from 29% a year earlier. Being absent can mean never making the shortlist.
G2, 1,076 buyers
76% → 38%
Ranking still matters, but less. The share of AI Overview citations that also rank in Google’s top 10 fell from 76% to 38% in eight months.
Ahrefs, 863,000 SERPs

That is ten. The report has the rest, with the studies behind each one.

The spine of the report

Proven vs. hype, side by side

What the evidence supports, and what is marketed hard but does not hold up. Every tactic in the full report carries an evidence grade — here is a taste of each column.

Evidence supports it — do this

Be crawlable and indexed
Proven
Google’s confirmed hard floor: no index, no citation. The highest-ROI 15 minutes on the list, because it is free.
Answer-first content
Proven
About 44% of AI citations come from a page’s first third. Put the answer in the opening paragraph.
Off-site brand mentions and reviews
Proven correlation
Correlate about 3x more strongly with AI visibility than backlinks. A few genuine mentions likely beat ten more of your own pages.
Question-style headings, real FAQs
Promising
Question-phrased H2s get cited about twice as often as statement headings.
Rank well, broadly
Proven — weakening
Still the densest doorway — but top-10 overlap fell 76% to 38%, so rank across many phrasings, not just #1.

Marketed hard — the evidence says no

Schema markup as an “AI-citation lever”
Debunked
Controlled test found no lift. Keep schema for classic rich results; stop paying for it as an AI hack.
llms.txt files
Debunked
97% got zero crawler requests; Google ignores them. A solution with no consumers.
“61% of AI Overviews use bulleted lists”
Fabricated
Follow the citation and neither cited study contains “61%” or “12%” — a real ~40% “has a list” figure mutated into an invented split.
“Verified Bing Places = 3–5x higher ChatGPT ranking”
Fabricated
An unbroken wall of SEO blogs citing each other. Zero primary sources, zero methodology, no OpenAI documentation.
Guaranteed AI rankings, “converts 23x better”
Debunked
Answers change every run; the multipliers are n=1 vendor anecdotes. The FTC began prosecuting “AI-washing” in 2026.

That is 10 rows. The full verdict tables — 40-plus graded tactics, plus the fabricated stats dissected as step-by-step detective stories — are in the report.

Free this week

Three things you can do this week — free

Most of what is sold as “AI SEO” is slow, expensive, or unproven. These are not. Each is cheap, backed by real data, and doable by you or your web person in an afternoon.

1

Let the right AI crawlers through

Open yourdomain.com/robots.txt and confirm you have not accidentally blocked the bots that feed AI answers: Googlebot, GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Applebot. You cannot be cited if you cannot be fetched.

Effort: 15 minutes
2

Put the answer first

Rewrite each important page so the first sentence under every heading is a complete, standalone answer — not “as mentioned above, it varies.” 44.2% of AI citations come from the first third of a page, and it reads better for humans too.

Effort: 2–4 hours, free
3

Complete your Google Business Profile and gather reviews

If you serve a local or regional market, fill out every GBP field and build a steady flow of real reviews. It is the closest thing to a sure bet in the whole report — Gemini pulls straight from Google Maps — and one study found brands with no active review profile were cited about 1% of the time versus 75% for those with a managed profile.

Effort: 1–2 hours, then ongoing

Five more quick wins — plus a “skip these this week” list of what not to waste money on — are in the report’s checklist.

Methodology

How this report was researched

Here is how the report was put together, and what was cut along the way.

631
primary sources reviewed — company filings, academic papers, first-party studies, regulator records
36
research areas — each treated as its own specific question, not a footnote
narrowed down
394
initial findings
224
consolidated claims
151
claims fact-checked
99
Confirmed — the source says what the claim says
39
Corrected — a figure, date, or scope was fixed to match the source
11
Downgraded — real, but hedged to what the evidence actually supports
2
Rejected as fabricated — cut, and written up as a worked example

Five contested questions, where credible sources genuinely disagreed, were resolved with dedicated follow-up research rather than a convenient pick. Where sources still disagree, we say so — which is why several chapters give you a range instead of one tidy number.

Where the evidence is thin, the report says so instead of rounding up. Every chapter ends with its full source list, so you can check any number yourself.

The full contents

What’s inside

13 chapters~27,000 words65 pages

Written in plain English. Every chapter ends with its sources.

01
Executive summary
The ten headline findings and the report’s argument in one page.
02
The 2026 AI search landscape
What actually changed this year, and what did not: zero-click, the CTR damage, and the AI-traffic mirage.
03
How each AI engine picks its sources
The crawler cheat sheet: what fetches, indexes, and cites you, engine by engine.
04
What actually works: the evidence
Every tactic graded Proven, Promising, or Debunked, with the study behind it.
05
The hype list
The “don’t bother” list, plus fabricated stats taken apart as detective stories.
06
Quick wins
The cheap, mostly-free checklist you can run this week.
07
The complex layer
The slow, compounding work: entities, digital PR, and review platforms.
08
SEO is still the foundation
Why “SEO is dead” and “GEO is a new discipline” are both wrong — and both cost you money.
09
The local and B2B service playbook
Two separate games: someone near you needs a job, and someone is choosing a vendor. Includes a German / DACH note.
10
Measuring AI visibility without wasting money
The free measurement stack, and the dashboard scores that mean nothing.
11
The grounded 2026 outlook
What is durable, what is likely, and what nobody honestly knows yet.
12
How this report was researched
The full verification process, the verdicts on every claim, and the report’s own limitations.
13
About WilkiLeads
Who wrote it, and the €99 next step.

Free PDF · instant download · no spam.

The person behind it

Who wrote this

WilkiLeads is a one-person web-design and SEO consultancy run by Eric Wilkinson, building and improving websites for small businesses across several countries and helping them get found — in ordinary search and, increasingly, in the AI answers this report is about. You deal with Eric directly, with the same skepticism you have just read: he would rather tell you a tactic is not proven than sell you a package around it.

If the report is useful and you want the next step, the most direct one is the Complete SEO Action Plan — a fixed-price €99 audit that turns these fundamentals into a specific, prioritized to-do list for your site: what is blocking you, what to fix first, what to ignore. No retainer, no lock-in, no guaranteed-ranking promises — you will know why those are not real.

See the €99 Complete SEO Action Plan → Or book a free call →

Get the honest version of AI search in 2026.

The full 65-page report is free. Read it before you spend money on “AI SEO”.

Prefer to talk it through? Book a free 30-min call →

Free PDF · instant download · no spam, unsubscribe in one click.

Questions

Before you download

Is it really free?+
Yes. No payment, no card, no trial. Enter your email and the PDF downloads instantly.
What happens with my email?+
It is used to send you the report and, now and then, a useful WilkiLeads update you can unsubscribe from in one click. It is never sold or shared. That is it. Details: privacy policy.
Who is it for?+
B2B and small-business owners, local service firms, independent shops — anyone who wants the truth about AI search without the agency hype. No SEO background needed; it is written in plain English for the person running the business.
How current is the data?+
It is a snapshot of July 2026. This field moves monthly, so the report also tells you which numbers to re-check and how often, instead of pretending any figure is permanent.