Hi everyone, this is Neo.

Today’s article features someone special: Jason Shafton, who was at Google early on helping scale the Google Ads platform — billions of dollars in advertiser budgets flowed through systems his team ran. He now runs his own growth consultancy (Winston Francois) and spent the past year running AI search visibility projects for a dozen or so VC/PE-backed startups.

He wrote a long piece for Search Engine Journal with a long title and one core sentence: if I were rebuilding a search strategy today, this is how I’d build it — applying the same discipline I used on Google Ads to AI search.

Why is this worth ten minutes of your time? Because he drops a number that sends a chill down traditional SEO practitioners’ spines — and it comes from Semrush’s official research:

Nearly 90% of the pages ChatGPT cites rank 21st or lower in traditional search results.

Read that twice. Most of the sources AI assistants trust aren’t first-page winners at all. The “get into the top 3” instinct that drove the past two decades is nearly invisible in this new channel. For independent store owners who treat “rank on page one” as an article of faith, that’s both bad news (the old playbook is losing its power) and good news (there’s still a window for newcomers).

---

1. Three iron rules I learned at Google that still hold today

Jason says the difference between winners and losers in the Google Ads era was rarely intelligence — it was these three disciplines:

1. Platforms reward the signals they can quantify, not your effort

“Advertisers have never once succeeded in buying their way through a bad Quality Score,” he says. “Winning accounts did one thing: give the system unambiguous evidence of relevance, then let the machines handle the rest.”

2. Budget follows evidence — and it almost never does

“Every account has a line item that exists only because it existed last quarter. The best operators hunt down those habitual spends for sport.”

3. Every platform shift pays the people who adapt first

Broad match, mobile, smart bidding — each one was declared “the death of the channel,” and each one made serious players serious money. Because most people wait for the “best practices playbook” to appear before moving — and by the time it does, the money’s already been made.

Neo’s take:

I’m 100% on board with all three. Especially the third — Chinese independent store owners share a common flaw: they wait for “certainty” before acting. Waiting for Google to publish docs, waiting for influencers to publish tutorials, waiting for peers to figure it out… by the time all that lands, the competition is already white-hot. AI search is in the classic “rules just changed, no playbook yet” phase — and that’s exactly the window of opportunity.

---

2. The core shift: AI search isn’t changing “how results rank” — it’s changing “what a result is”

Jason’s read on AI search is worth savoring:

AI search is the biggest of all these shifts — because it doesn’t change how results are ranked, it changes what “a result” is.

In traditional search, a result is “a list of links.” In AI search, a result is a synthesized answer: the assistant reads a dozen sources, writes a response, and names a few brands. “The answer is the result; your link is optional.”

Your buyers are now asking ChatGPT, Perplexity, Gemini, and Google AI Mode the questions they used to type into a search box. The assistant reads dozens of sources, assembles an answer, and names a few brands — if your brand isn’t on the named list, you don’t exist in that customer journey.

He also shared two real audit cases with a striking contrast:

• Client A: middling Google rankings, but a strong footprint on Reddit and industry media → AI assistants cite them consistently, far more often than their rankings would suggest
• Client B: holding several page-one spots in the core category → zero AI answers for core purchase-intent questions

Neo’s take:

These two cases are living examples of “90% of cited pages rank 21st or worse.” The battlefield of AI search isn’t your homepage at all — it’s the entire web’s worth of “how other people talk about you.” A lot of us are still grinding on homepage rankings, when in the AI-citation dimension, ten thousand words from your own site mean less than one sentence from a third party.

---

3. The 6 visibility signals (the important part — bookmark this)

When Jason’s team audits clients, they find these six signals explain most of the variance in AI visibility. Notice — most of these six aren’t on your own website:

#	Signal	Key point
1	Brand Authority	AI assistants favor brands that get “named” by others on credible sources. Your own blog barely counts; third-party mentions are what matter.
2	Content Freshness	Pages untouched for two years get skipped. Visible, dated, real updates matter more in AI search than in traditional SEO.
3	Entity Recognition	These systems think in entities: who you are, what you’re known for, which real people in your company are recognized as experts in the field.
4	Extraction-Ready Structure	Clear heading hierarchy, answers right at the top of the page, schema where appropriate. As Danny Sullivan said: structured data “supports understanding,” it doesn’t “guarantee indexing.”
5	Cross-Platform Consensus	Reddit posts, YouTube videos, review sites, and media coverage all describe you the same way — assistants read that consistency as trust. This is the hardest to do and the hardest to copy.
6	Earned Media	Third-party coverage you didn’t pay for and didn’t write yourself carries remarkably high weight when it gets cited.

Jason’s own summary: “Notice the composition of this list — most of it is brand and PR work, just run with performance-marketing discipline. If your AI search plan is confined to your own website, you’re working on the smallest part of the whole system.”

Neo’s take:

Of the six signals, #1, #5, and #6 are essentially three facets of the same thing: third-party, consistent, earned descriptions. This matches exactly what I’ve been telling readers — in the AI era, an independent store’s moat is shifting from “what you own” to “how others describe you.”

For small and mid-size sellers, signal #5 (cross-platform consensus) is actually the best one to attack first: it doesn’t require media relationships — just unify how your website, marketplace store, social media, YouTube, and G2/Trustpilot profiles describe you. Get “one voice” locked in first, then chase coverage.

---

4. The 90-day sprint: a three-phase roadmap

Jason’s framework is his standard growth-audit structure, applied to AI visibility in three phases:

Phase 1 (Weeks 1-2): baseline diagnosis — the “20-run test”

You can do this this afternoon:

1. Write 5 questions real buyers would ask, with purchase intent behind them — e.g. “best [category] for [use case]”, “my product vs. competitor”
2. Run them through ChatGPT, Perplexity, Gemini, and Google AI Mode — 20 runs total
3. For each run, record: named / cited a link / completely absent

The bar: if you show up in fewer than 12 of the 20 runs, you have a “citation problem” — and a clear direction for the next quarter of work.

Then make the baseline sustainable: use tools like Profound, Otterly.AI, or Peec AI to track citation changes over the long term.

Phase 2 (Weeks 3-6): run one “small experiment” per channel

One thing per channel, something whose results you can see within 30 days:

• Website: rewrite 10 conversion pages into “extraction-ready” format — direct answers at the top, comparison tables added
• Reddit: build a genuine presence in the 2 subreddits where your buyers hang out (answer questions, don’t advertise)
• Reviews: push for new reviews on G2 or the platform your category trusts
• Media: pitch 2 data stories to industry media (your data, your case studies)

Where does the money come from? Cut the budget line that “nobody has questioned in six months” — Jason says every marketing org has one.

Phase 3 (Weeks 7-12): the compounding loop

• Kill the actions that aren’t driving citations
• Double down on the channels that work
• Build a weekly ritual: the same 20-run test, the same prompt set, tracked against the baseline
• Give it a named owner — in most teams, that should be the strategy lead

Jason’s real-world results: this spring, a B2B client went from 4/20 to 13/20 on the 20-run test in one quarter, with the biggest jump coming within two weeks of their first batch of Reddit answers going live.

Neo’s take:

What wins me over about this plan is its falsifiability — every step has a numeric baseline, and you know within 30 days whether it’s working. Compared to the common approach of “make content for three months and see if traffic moved,” this method turns AI visibility from mysticism into a manageable performance metric.

A localization tip for independent store owners: you can swap the Reddit piece for the platform that matches your target market (Reddit for US/EU B2C; add LinkedIn and industry forums for B2B; if you’re selling into Southeast Asia, switch to the community platforms there). The methodology is universal; pick channels by your market.

---

5. Closing quote: why this is “media-buying discipline,” applied one level down

At the end of his article, Jason says something I think deserves a spot on your wall:

This sounds like applying media-buying discipline to a channel you can’t buy — and that’s exactly what it is. You can’t bid your way into AI answers. You earn your position with evidence, and evidence requires reps.

He also offers a conceptual translation I really like: “Citation Share” is the AI-era “Impression Share” — back in the Google Ads days we watched impression share; now you should watch: of the purchase-intent questions that matter to you, what percentage of AI answers name you?

Semrush’s research also quantifies the long-term value of this channel: AI search visitors convert at 4.4x the value of traditional search visitors, on average — because AI has already done the homework, visitors arrive with a clear decision in mind, and conversion rates are naturally higher. And by Semrush’s projection, by early 2028, traffic from AI search could surpass traditional search.

So go run your first 20-run test now — it takes 30 minutes, and it will change the mood of your next marketing meeting.

---