❌

Vue normale

Reçu — 28 octobre 2026 ⏭ Occam’s Razor by Avinash Kaushik

SEO + AEO = Organic Answers Optimization [OAO] Analytics

It is time for a new acronym to enter the marketing lexicon: OAO.

We have lived in the Search Engine Optimization (SEO) world for the longest time – it remains critically relevant. Many of us have embraced the new demands of Answer Engine Optimization (AEO). [I’m sorry, I’m not a fan of Generative Engine Optimization (GEO) because that honors the technology and tech changes all the time.]

Organic Answers Optimization, bringing the two worlds together… Because they really are one thing now.

Big brother SEO is still dominant – most traffic, most revenue. Little brother AEO has “stolen” “market share” in the consumer influencing game (no need to come to our sites ☹️), has started to drive modest scale in traffic and a surprising scale in profit.

Practitioners in each space fight surprisingly passionately to preserve their respective SEO and AEO domains. It. Is. Ok.

For me, the business person, it is now just one thing:

Work to influence a machine to recommend my brand, and convert that influence into relevant traffic.

 
My SEO work drives AEO, and AEO drives SEO. The brothers (or sisters, if you prefer), united in strategy, execution, tactics, impact, and analytics.

Different “machines,” one quest: Organic Answers Optimization!

It makes it easier to make peace with this:

Your SEO traffic is likely down 20%-30% YOY (as I predicted!).
Your AEO traffic is up from 0% to 2% (as I predicted).
Overall “organic” revenue is down (unless you did most of the 12 SEO and 8 PPC recommendations, in which case Rev up ~10%).

I want to bring these two together, from a business perspective, because “organic” deserves extra special attention. Even bigger declines in traffic are on their way.

This post was first published as TMAI Premium #531. Every week, serious Marketing & Analytics professionals get strategic frameworks (how to think) and actionable advice (how to do) – proven at the world’s largest companies. Subscribe today.

Context for Change.

Around this time last year, I wrote a six-part (!) series on AEO, AEA (Analytics), and a three-part series on how to identify the upcoming losses, take actions for recovery, and compute the financial impact.

[Annual Premium Subscribers: I have converted all this into an updated 114-page Answer Engine Playbook! If you would like the PDF version of the mini-book, please email me.]

In a keynote I subsequently delivered, this picture captured the evolution:

Consumer seeking behavior change: From searching to getting answers to getting AI agents to complete the tasks

Let’s say I have an upcoming trip to Prague.

Search

Old-school search would have me going to Big Google and searching for “lightweight carry-on luggage.” I’ve been patterned to use keywords to get back three paid links and ten organic links, and then head off on a research mission to find the right suitcase.

Most of us do this (a billion plus).

Answer.

Along with approx. 50% of GenZ who use an LLM every day, I stopped doing searching 18 months ago. I now pop open ChatGPT and type “ I need a light carry-on for a ten-day trip, with a small European airline connection, for under $250. I also need a backpack that I can stuff with extra clothes and a laptop, to go with the bag that the small EU airline will allow me to carry on.” Astra (my GPT model) is now off to the races, and 17 seconds later I have my answer. No research, no sweating.

A few hundred million of us, skewing young, are doing this.

[Bag: Travelpro Maxlite International Carry-on Spinner. Backpack: CabinZero Classic Tech 28L. Both look amazing. I’d never heard of CabinZero: the discovery power of an LLM.]

Complete.

Increasingly, I skip the buying part customers doent do it. Same context-rich query as above, but I might add “ …buy the first recommendation for each, confirm it will arrive by Oct 30th, from a site with a clear returns policy.” Often, the purchase shows up. Sometimes the site blocks my Agent from buying, so it adds to cart and sends me the link.

1% to 2% of us are doing this.

The world is moving right, because why would a human need to spend time hunting and gathering, the meal can arrive all caught, plucked, seasoned, and cooked?*

With Dots and Muse, the pace will pick up.

*Some people find joy in shopping; they will continue to have this option!

Why One OAO Scorecard.

Your customer does not care if they used Old Google or used an Answer Engine. For them, they were just seeking, and they found what they wanted.

[Per SparkToro, approx. two-thirds of US Google searches end without a click.]

Your CFO does not care about Old Google or Claude or Qwen. They care that we are chosen and get revenue.

[Per Adobe, with a trillion US retail visits analyzed, LLM-driven visits deliver 53% more revenue per visit than other channels. Per Shopify, LLM-driven visits convert approx. 50% better than Organic Search.]

Your Marketing team does care because you have an SEO team/agency and, if you read my mini-book, have a raft of AEO initiatives in your Content team. And these are silos today because the work used to be different. SEO, which you already know all about, had an emphasis on 1P and text plus images. AEO was built on that strong foundation, and added an emphasis on 3P and video plus audio.

As customers delegate more, a visit to your website becomes a less complete record of your influence. Hence, pulling it all into one scorecard will provide a fuller view of influence, and create incentives for your silos to break.

One more reason… For now, the first strong commercial results from your “AEO” efforts will flow in through your “SEO” traffic. Unification helps here too.

Meet Kettle & Loom.

Learning is better when we have an actual use case. Kettle & Loom is a made-up home goods retailer. K&L’s CMO was given a report with three numbers:

1. Search Console Clicks: -14%
2. Agency Visibility Score: +9%
3. Organic Revenue: +2%

Is that good? Bad? Yes. 😊 Let’s learn.

OAO Analytics Methodologies.

For our unified scorecard we need metrics. BUT. Methodologies trump metrics!

Let’s do that first.

There is no one ring to rule them all. At this stage, we will have to pick and choose from a basket – and there is a lot of “variable” quality out there.

Your owned systems (GA, CRM, ERP) describe visits, behavior, sales and profit, mostly to entirely accurately. Search Console describes observed Google Search performance – often aggregated or hidden for privacy reasons. Competitive tools estimate things about other sites that your analytics can’t see – hence, they’re best for comparison, never for raw numbers.

Organic Answer Optimization: Analytics Methodologies | Good, no good.

It is important to reflect on methodologies carefully because it really is a case of GIGO. And, in our world of immense scale, it is more like garbage in, landfill out.

It is particularly critical for Answer Engine Analytics tools as they are so new. Investigate how they collect data. [Premium Subscribers please see: TMAI #472: AEA Methodologies & KPIs. If you can’t find it, please email me.]

There is a lot of snake oil out there. Ex: Some vendors claim, or STRONGLY imply, that they have full access to actual prompts typed by users of LLMs. That is simply a lie.

Panels of a few mil can be somewhat useful at extremely aggregated levels. Ex: Industry. Even by category in an industry, it falls apart. For a single brand in a category in an industry, it is just crap.

Even a large panel will see 0.01%, or significantly less, of ChatGPT’s July 2025 publicly stated 2.5 billion prompts. For every MILLION times a question is asked about your category, a vendor claiming “tens of millions of prompts captured” will catch about 100, and only single digits, at best, will be about you. Worse news? The monthly “search volume” of any one prompt is effectively one.

One day we might have “Claude Webmaster Tools,” or “OpenAI Seeking Console,” but until then… no dice.

SEO also has issues we have learned to be careful about. Marketers constantly mistake estimated traffic, authority scores, and visibility indexes – all proprietary vendor guesses – for Google’s actual ranking mechanics (falling victim to Goodhart’s Law).

K&L’s Agency Visibility Score of +9%? Their vendor’s model of rankings * estimated clicks. Some value in comparing against direct competition. It is not proof of any extra dollars.

Everything in AEO for now, at least, is a “long tail query,” but we also don’t get such queries from Google – these hidden rare queries add up to nearly half of all clicks (Ahrefs). 3P SEO tools that “model” this long tail data should be viewed with suspicion. Google now mixes AI Mode questions into the same report, unlabeled (and rare, hence, more likely than not, hidden).

K&L’s -14%? Real clicks. But brand and non-brand are blended. AI Mode sits inside, unlabeled, and nearly half the queries are hidden. That number is a mush of all that.

Just because a source is not perfect does not immediately mean it is useless. Understand. Use metrics from the methodology at the signal strength it offers.

Humility Note: Organic Answers Optimization is funneling demand that in large part was created by brand advertising, distribution, seasonality, word of mouth, yada, yada, yada.

K&L’s +2% organic revenue? 50% of it comes from brand searches, and most likely from non-S&A EO initiatives.

Organic Answers Optimization CxO KPIs.

KPIs are metrics that directly impact the bottom line. Typically, they are for people who fund things. They tend to move over months and quarters. They are most influential – or you are doing them right – when they have pre-set targets.

For organic traffic, across all machines, from an ocean of metrics and fuddy counts, I recommend three KPIs (you should rarely have more than three):

KPI #1: Non-Brand Organic Search Profit.

Measures profit from orders (or leads) that began with an unpaid search visit from someone who did not search for your brand/product/service name.

Best source of this data: GA4 by landing page + Search Console branded filter + GA4 AI Assistant channel. It is imperfect, but as good as it gets.

[Premium Bonus: Please see TMAI #521: GA 2026: Unlocking New Value re how to improve GA’s ability to report LLM traffic more accurately using my AI True Traffic Total Kit.]

Why insightful: Following my humility note above, the real job of OAO is to help find and deliver people who didn’t know you – brand demand has extremely low incrementality.

Action recommended: Set targets for this number, then fund SEO initiatives against it (to mostly slow down declines in Organic Search traffic – using, among other things, ideas in my Loss/Recovery spreadsheet model).

Pro-tip: You can translate KPI #1 into Rental Cost. Non-brand organic clicks * what you’d pay per click for the same queries = What it would cost you to rent an asset you own!

KPI #2: Answer Engine Profit.

What it says on the tin.

Best source of this data: Site analytics tools, AI Assistant channel. This is incomplete (mid-Oct ChatGPT decided to change referral strings), and a bunch of it lands in Direct. But. We have to start somewhere.

Why insightful: Is the trend Adobe and Shopify have identified, higher conversion rates and AOV, happening in your case? Profit numbers bring focus.

Action recommended: Use it to gain investments in content (especially video, audio, especially brand, especially 3P). This data can help you persuade your sometimes stubborn, old-school Tech teams to focus on technical AEO, unblock robots and Agents.

Pro-tip: It is really hard to quantify, but most AEO content investments will initially return as a reduction in SEO declines (hence more SEO revenue).

Organic Answer Optimization: Best KPIs, Metrics

KPI #3: AI Brand Score.

An approximation for how often AEs recommend you, and how prominently, when the question doesn’t include your name. It combines Visibility Score with Average Position.

Best source of this data: From tools like Evertune*, a probabilistic answer gained from repeated prompts to identify a distribution of answers the machines provide. 100 = your brand named first, every time. What’s yours?

What it isn’t: It doesn’t claim to know what your customers type. Nobody knows that.

Why insightful: The number closest to the Most Important Marketing KPI ever: Baseline for Unaided Brand Awareness. You can also think of it as your share of an LLM’s shelf.

Action recommended: I believe the insurance against all the changes coming is my strategy of more content, more places, more often (details in my AEO mini-book). AI Brand Score measures progress against this faith in future influence and revenue. Hence, proof for your CFO that content investments are worthy.

Pro-tip: Tag every major model release on your AI Brand Score chart. A new model can swing you ten points instantly – without changing anything fundamental. Don’t take credit, or blame, for the AI’s homework.

*A long time ago, I made a tiny, tiny, tiny angel investment in Evertune because the team’s awesome.

Let us reflect on K&L’s measurement and apply the lessons above…

Measurement Quality Assessment: Metrics, KPIs.

The Story Continues in Parts 2, 3.

K&L’s CFO will ask the obvious next question: why is non-brand profit down? Did K&L lose rankings? Did Google keep the click? Or did customers take their questions to LLMs and never click at all?

TMAI Premium #532 built a simple triage process, entirely from data K&L already own. This quickly helps us identify challenges/opportunities each month. The next step is to drill down to diagnostic metrics and influencing variables that K&L’s agency should be acting on. For larger brands, TMAI #533 also pulled out of SEO the contribution AEO makes to reducing declines or increasing growth.

The journey ends with your awesome OAO Analytics dashboard:

Organic Answers Optimization Analytics Dashboard

The mini-book worth of stories put together reduces faith required in my OAO strategy of more content, more places, more often – by providing data.

This is the

Premium Subscribers also received an Excel workbook to use across VPs, Directors and agency peers to activate SEO+AEO=OAO revolution. To get a copy sign up here, and reply to the welcome email.

Bottom line.

As humans move from Search to Answer to Complete, the cascading impact on businesses is immense. It’s hard to appreciate for now, because Search remains dominant – especially for older demographics (who have all the money!).

My hope with my little new acronym, OAO, is to encourage a focus on change to ensure change does not catch you by surprise – delivering a business shock.

Faith + Data will prevent surprises.

Carpe diem.

Google’s Statistics: Heads They Win, Tails You Lose.

14 septembre 2026 à 10:14

Google Brand Lift & Conversion Lift studies use the same underlying idea of statistical certainty. But certainty is not the same thing as answering the business question that matters most: if you spend the money again, how likely are you to reproduce the result?

Let’s start with a phrase that causes me to cringe every time I hear it: “directional results.”

It sounds sophisticated. Nuanced. Even reasonable. But what does it actually mean?

Answer: Just about anything you want it to mean!

I was reminded of this while reading Google’s official guidance on how to interpret Google/YouTube’s Brand Lift, Search Lift, and Conversion Lift studies.

Google Brand Lift and Conversion Lift certainty guidance

It made me pause. Then sad. Read the last one carefully.

Google says: ≥90% is “very good chance.” 70%-90% is “good chance.” 50%-70% is “moderate chance.” Your ads probably did something good.

The last layer is, paraphrasing: Hey, the ads might not have sucked, spend some more and let’s see what happens.

Google also says in the article: Studies at 50%+ can provide valuable, directional insights, and explicitly tells advertisers in the 50%-70% band to use the results directionally.

If you pause and reflect on this carefully caveated article, you’ll see this is a case of: Heads Google wins, tails you lose.

This post was first published as TMAI Premium #523. My weekly newsletter cuts through the noise: Strategic frameworks, actionable advice, zero fluff – built on decades of doing, not just advising. For serious Marketing and Analytics professionals. Subscribe today.

Whose Success Matters More?

To bring this lesson home, let’s assume I ran a campaign on YouTube.

  • Investment: $5 million
  • Brand KPI: Consideration
  • YouTube-reported BLS result: +3-point lift
  • YouTube-reported Certainty: 70%
  •  
    Google calls 70%-90% Certainty a “good chance.” It also says studies at 50%+ can provide “valuable, directional insights.”

    What. The. Heck. Does. Directionally mean?

  • Do I spend another $5 million on another YouTube campaign?
  • Do I spend less and do something different, in a different place?
  • Do I change the creative or media tactics?
  • Do I shift the money to Meta that presented a new exciting idea yesterday?
  • Do I run another “test” (and to test what)?
  •  
    From Google’s view, the interpretation is that there is a “good chance the results were caused by your ads.” But as a Senior Director of Brand Media, is a “good chance” good enough for me to report to my CFO and ask for $5 mil more?

    Two Parties. Two VERY Different Consequences.

    Party One: Google: It sells advertising. If a study contains weak evidence that advertising could possibly have worked, Google has every incentive not to throw it away. Perhaps, and that’s the operative word, perhaps there is something useful to learn. Fair enough.

    Party Two: You: You own the Marketing budget. Your job is not to find something interesting in the data. Your job is to answer: IS the evidence in YouTube’s BLS strong enough that I should risk another $5 million of my company’s money on the conclusion that the first $5 million worked?

    Here, I made a little table to outline incentives and implications:

    Google Math's world view vs. CFO Math world view.

    It is impossible to know if Google’s wording on its sites, in its client decks, is intentionally self-serving. It does not matter. The outcome is self-serving.

    Google does not bear the consequences of an optimistic interpretation. Your company does. Google also, to their credit, has a drive-by line that says: Guys, interpret results based on your business needs and risk tolerance.

    Since I am bearing the risk, for consequential business decisions I use 90% Certainty as the floor. It isn’t magic; it is simply the minimum level of evidence I’m willing to accept before telling a CMO or CFO: “This worked. Spend again.”

    [Premium Subscribers: Refer to TMAI #298: Smarter Statistical Significance Reporting. It includes a one-pager to standardize stat sig reporting across your team, agency, vendors. If you can’t find it, please email me.]

    Google’s emphasis is on “moderate chance,” “good chance,” “directionally.” These are comforting, decision-sounding words that obfuscate evidence more than illuminate it. The language is for sure carefully caveated, but it ignores the immense risk and cost it intentionally or unintentionally shifts onto Google’s clients.

    Let me prove it to you.

    Google Brand Lift and Conversion Lift: Certainty vs. Replication Risk

    Rather than discuss p-values associated with “good chance,” “moderate chance,” “use these results directionally,” let’s focus on the question your CMO and CFO care about:

    If we ran this campaign again with another $5m budget, how likely is it that we will get a result strong enough to confidently say: “Round two worked. Spend again.”

    You’ll recall YouTube reported a +3-point lift in Consideration at 70% Certainty.

    While they might not use the technical phrase, your CMO and CFO are essentially asking: What’s the Replication Risk?

    [For this thought experiment, we’ll assume, the second $5m will face the same underlying conditions as the first $5m. Same platform (doh!), audience, seasonal factors, promotions, creative, sample size, etc.]

    To compute Replication Risk, I’ll use a simple Bayesian replication model inspired by biostatistician Steven Goodman. It helps translate Google’s abstract “Certainty” into a much more practical question: What might happen if we measure the same thing again?

    YouTube’s BLS estimated a +3-point lift in Consideration. But +3 is an estimate; rather than pretending it is precise, the Bayesian approach carries that uncertainty forward into the next measurement.

    As business owners, it is our money on the line. We are applying this simple logic:

    (Uncertainty from what the true effect might be)
    combined with
    (Uncertainty from measuring it again)
    = How confident should we be that we will see similar results?

    Let’s do some math!

    Three computations for the Business Owner lens:

    1. Repeat Shows Positive Lift: The probability thta an equivalent repeat produces any positive lift, even a tiny, tiny, tiny +0.01 points – an extremely low bar.

    2. Repeat Reaches ≥90% Certainty: The probability that an equivalent repeat produces ≥90% Certainty – my minimum standard for evidence strong enough for us to spend again.

    3. Replication Risk: Simply the opposite of #1. The probability that the repeat falls short of the ≥90% standard.

    Human language: What are the chances I spend another $5m, and still don’t get a result with certainty for me tell my CMO or CFO, “This worked. Spend again.”

    Punch to the gut: If your Replication Risk is 70%… There is only a 30% chance that spending another $5m will result in you telling your CMO “This worked.” #killmefirst

    Google vs. You: Google calls 70% Certainty a “good chance.” The replication math shows the “much, much less than good chance” you’ll take in round two.

    Here’s the model using a simple Bayesian replication model inspired by Mr. Goodman:

    Results replication risk calculations.

    [Pedantic Note: Google says displayed Certainty is rounded down in 5-point increments. So a displayed 70% can represent 70% to just under 75% and hence the exact replication % can move a few points. For simplicity, I treat Google’s displayed Certainty as exact.]

    YouTube had reported 70% “Certainty,” follow the row above… That “good chance” Google label translates into:

    1. Just a 65% (!) chance of any Positive Lift, with a

    2. A puny 30% chance of hitting high certainty,

    3. when repeating the same spend ($5m) under the same conditions.

    As a business owner, would you feel confident saying yes, let’s spend on YT again?

    Forget Google for a second, take me: I recommend a minimum of 90% Certainty for a business decision. That comes with a Replication Risk of 50% (!!) and Repeat Shows Positive Lift of just 82%. Too uncomfortable in many cases (what if it was a $10m question, would you take 50% and 82%?)

    Now Alex and Juliette understand why I so often push for 95% Certainty. When you spend as much money as we do on YouTube… At least get the Replication Risk down to 40%, increase the Repeat Shows Positive Lift to 88% and increase the odds more in the company’s favor.

    So… Are results below 90% Certainty useless?

    If you have real money riding on it, look at column 3, look at column 5 in the table, see if you want those odds. If you have little to no money riding on it, sure – go try a new hypothesis, design a different/better test, give additional money to Google/YT/Meta/TT, remeasure.

    A bonus relevant lesson from “Beyond Power Calculations: Assessing Type S (sign) and Type M (magnitude) Errors”: Weak evidence can hurt twice: The evidence may not survive (70% in our case), and the apparent impact might be overstated (+3 in our case).

    You’ve been warned.

    Premium Subscriber Bonus: A Working Model.

    I want you to be able to see all the formulas and math I’ve used in my computation of Replication Risk.

    I want you to be able to test lift’s Certainty AND Magnitude (Repeat Shows Positive).

    I want you to be able to type in the results of your Brand Lift, Search Lift, and Conversion Lift studies and calculate your own truth (before you put your career on the line).

    To enable all this, I’ve built a working model with Read Me + Methodology, Calculator, Risk Table, and Calculate Your Truth tabs. It is a bonus for TMAI Premium subscribers; please just email me for lift_replication_risk_model.xlsx.

    Bottom line.

    Google and you are sitting on opposite sides of a consequential decision.

    Google sells advertising. You are spending money.

    Google might reasonably find value in evidence that is merely suggestive. See the first table. You have to decide whether that evidence is sufficient and measured credibly enough to risk the next tranche of your budget.

    Don’t let an advertising platform’s definition of useful information become your company’s definition of sufficient evidence.

    Stop outsourcing your tolerance for risk to the company selling you ads or the Agency whose fees are tied to the size of your spend.

    Carpe diem.

    PS: We have used Google’s specific table and specific guidance today. Replication risk applies just as much to Meta, TikTok, and confident calcs your Agency is sending you. Please check just as carefully if they are solving for themselves or you.

    ❌