|
HQ DIGITAL · WEEKLY NEWSLETTER
Ground Truth
|
ISSUE 05
13 MAY 2026
|
|
|
THIS WEEK
Position 1 on 14 keywords.
Traffic still collapsed.
A teardown of the 5-layer stack we built above SEO to recover what AI Overviews and answer engines took away.
|
|
THE SETUP
Last month I sat with the founder of an Indian D2C skincare brand. NDA, so the name stays out, but the work is real and the numbers are unmodified. Vitamin C serum as hero SKU, six years of SEO investment, an in-house team of three.
His agency had him at position 1 for "vitamin C serum benefits" and 13 other primary keywords. Traffic had dropped from 4.1 lakh sessions a month to 1.6 lakh in 11 months. Same rankings. Less than half the traffic.
|
"But we are still number one."
— The founder, every time we discussed traffic
|
I pulled up the SERP. Position 1, yes. Above position 1 was a Google AI Overview that quoted three competitors and answered the full query in 70 words. Below it sat a People Also Ask box absorbing another 22 percent of attention. By the time a user reached his result, the question was already answered.
|
|
|
THE NEW BASELINE
This is not a one-off. It is the new baseline of organic search.
Two studies tracked this collapse in the open. They are the most-cited primary data on AI Overview impact in 2025-26.
|
|
|
|
PUBLIC CASUALTIES
|
HubSpot
|
−70 to 80%
|
|
Forbes
|
−50%
|
|
CNN
|
−27 to 38%
|
|
Wikipedia
|
−8%
|
Same rankings. Less than half the traffic. The click no longer follows the ranking.
|
|
|
THE FIX
The fix is not more SEO.
It is a new layer above SEO.
We call it the 5-Layer AEO Stack. Schema, Entity, Answer, Citation, Agent. Each layer decides whether you exist inside the AI answer or get filtered out.
|
|
|
CODE THAT TELLS MACHINES WHAT YOUR CONTENT MEANS
|
|
The brand's product pages had zero structured data. Clean HTML, but unmarked. To a machine, it was an unlabelled wall of text.
|
Where the brand started: clean HTML, zero structured data.
|
|
We added FAQPage, Product, BreadcrumbList, and Article schema across 240 pages. Two weeks of dev work. Two days of QA inside Google's Rich Results Test.
|
The three schema types deployed across all 240 product and content pages.
|
Validated inside Google Rich Results Test. Brand identifiers redacted under NDA.
|
|
90-DAY RESULT
60 → 3,400
AI referral traffic from ChatGPT and Perplexity. Zero new backlinks. Zero new content.
|
|
|
WHETHER MACHINES SEE YOUR BRAND AS A REAL THING
|
|
In August 2024, if you asked ChatGPT about the brand by name, it had no record. That is invisibility at the entity level. No matter how good the content is, AI cannot cite what it does not know exists.
Three actions: filed a Wikidata entry. Rebuilt the founder's LinkedIn with consistent organisation schema. Standardised brand mentions across PR, podcast appearances, and YouTube descriptions so spelling, city, and category descriptor were identical everywhere.
|
|
QUARTERLY RESULT
3× mentions
Brand mentions inside ChatGPT and Gemini answers tripled in the quarter following the Wikidata entry.
|
|
|
FIRST 50 WORDS MUST BE THE QUOTABLE ANSWER
|
|
The brand's blog had 60 long-form articles. The average first 50 words on every page was filler about how "skincare has become both an art and a science." Pure runway. AI engines extract in chunks of 40 to 80 words. There was nothing to extract. We rewrote the first 50 words of every page as a direct, quotable answer. Then the explanation followed.
|
Same page. Rewrite focused entirely on the first 50 words.
|
|
CITATION RATE
3.8% → 21.4%
Citation rate inside Perplexity across the tracked queries. 5.6× lift.
|
|
|
WHERE AI ENGINES SOURCE FROM
|
|
The brand was barely visible on Reddit. Quora answers in their category were 4 to 6 years old. No podcast appearances. Limited YouTube transcript footprint. Six-week effort: authentic Reddit comments where the founder answered ingredient questions, long-form Quora answers under the brand's authority, three founder podcast guest spots, transcripts cleaned and uploaded to YouTube.
|
|
MONTH 3 RESULT
8× mentions
Brand mentions inside Claude and Gemini grew 8×. Cited 3 times in the Perplexity source panel for the category query.
|
|
|
THE LAYER NOBODY IS DOING YET
|
|
Claude in Chrome, ChatGPT Operator, Manus. Browsing agents now navigate websites to complete real tasks. They book. They buy. They compare.
We tested the brand's site with Claude in Chrome. It could not complete checkout. The size selector required a hover state that did not register for the agent. Three fixes: removed hover dependency, added explicit ARIA labels, wrote an llms.txt at the domain root.
|
Live llms.txt at the client domain. Brand and product names blocked under NDA.
|
|
8-WEEK RESULT
0 → 240 orders
Agent-completed orders. Agentic traffic does not land on the home page. It goes deeper. The llms.txt is what makes agents find the right pages first.
|
|
|
THIS WEEK
Three things to do before next Wednesday
|
|
01
|
Audit your top 5 pages inside Google's Rich Results Test. Either you have FAQPage and Product schema, or you are invisible to AI extraction.
|
|
|
02
|
Ask ChatGPT, Perplexity, and Claude about your brand by name. The gap between what they say and what your sales page says is your roadmap.
|
|
|
03
|
Write an llms.txt for your domain. Twenty minutes of work. Massive future-proofing.
|
|
|
|
The full teardown with the schema templates, the llms.txt template, the entity-building checklist, and the Perplexity citation audit I run on every consulting account. Live workshop opens this week.
|
|
We do not optimise for clicks anymore.
We optimise to be the answer.
— Apurv Singh
HQ DIGITAL
|
|
Ground Truth
A WEEKLY NEWSLETTER BY HQ DIGITAL
You received this because you subscribed to Ground Truth at thehqdigital.com.
|
|