Chapter 09 · Stage 03
The Content Quality Gate
A page can keep its ranking and still lose its clicks to AI Overviews. Score how replaceable your pages are and rebuild the ones losing most.
- Rankings holding while CTR drops 35-72% across commercial page sets. AI Overviews assemble answers from commodity sources without needing any single one specifically.
- The five-dimension test (specificity, first-hand experience, viewpoint, LLM replaceability, stakes) scores every page 0-15. Most B2B content lands in the 5-8 range.
- The 5-8 trap is the most common failure mode: content that cites real sources and uses real numbers but lacks first-hand stakes.
- The swap test checks segment pages: if the construction page would serve a healthcare buyer just as well, the difference is cosmetic.
- Three moves: audit the worst-bleeding page first, build a sourcing pipeline before writing, stop publishing anything that fails the test.
Your flagship commercial page held position 1 on a category query and lost 62% of its clicks. The AI Overview cited eight competitors and did not cite you. Not ranked low. Not cited. The AIO did not need you.
This is the commodity content collapse, and it is the defining B2B SEO problem of 2026.
The mechanism
Section titled “The mechanism”AI Overviews assemble answers for generic commercial queries from multiple commodity sources without needing any single one specifically. Pages keep their rankings. Clicks go to the summary instead. The pattern across clients in Q1 2026: rankings holding or improving by roughly two positions on average, CTR down 35-72% across commercial page sets.
This is not a Core Update story. The pages did not get demoted. The SERP got rearranged around them. Pages that used to lose 20% CTR to featured snippets now lose 40-60% to AI Overviews. The losses compound with each update. Informational queries get eaten first because they are the most commodity-shaped. Commercial queries follow.
The standard advice is wrong. “Update your content,” “add EEAT signals,” “use more AI for scale” do not address the mechanism. These are responses to a ranking problem. This is not a ranking problem. It is a citation problem. The AI Overview does not need your page because your page contains nothing the overview cannot assemble from other sources.
What reverses it is a different kind of content.
The five-dimension test
Section titled “The five-dimension test”Every page gets scored on five dimensions, 0-3 each. Total range: 0-15. The test runs at score.baba-seo.com.
Before the full score, one pass-or-fail check: an intern with a good ChatGPT prompt cannot produce this. If the honest answer is that the intern could have written it, the score will confirm it and the page isn’t worth publishing. Anyone on your team can run that check on their own draft.
Specificity
Named instances, exact figures, verifiable particulars.
We helped a B2B SaaS company in a regulated category grow organic traffic from 12K to 52K monthly sessions over nine months, driven by 47 new MOFU pages.
Our clients see significant SEO improvements within a few months.
Find the most specific claim on the page. Could you swap the brand name and republish it on a competitor's site without changing anything? If yes, specificity is failing.
First-hand experience
Evidence the author actually did the thing being described. A first-hand account from a named person the author spoke to also counts: a customer, a practitioner, an in-house expert, quoted with the specific situation. The experience doesn’t have to be the author’s own; it has to be someone’s, on the record. “Experts say” or an unnamed “one customer told us” doesn’t count.
On a 2,000-page payments-company migration I led, we caught a canonical tag bug on day three that was silently redirecting roughly 400 product pages to the homepage.
SEO migrations require careful planning.
What proof exists on the page that the author did what they describe? Not that they read about it. Not that they know the theory. That they were in the room.
Viewpoint
Positions someone could disagree with.
Most SEO audits are theater. If the deliverable is a 60-page PDF, you hired the wrong person. A real audit names the three pages costing you money and tells you what to do about them by Friday.
Quality content matters more than quantity.
Is there a sentence on this page a competitor in the same industry would refuse to publish? If every competitor would happily sign their name to every sentence, the page has no viewpoint.
LLM replaceability
How much of the page a frontier LLM with no web access could reproduce from training data alone.
A piece built around a specific named deal closed last week, with a customer quote and the reasoning that drove the decision.
10 SEO tips for SaaS. An LLM with zero web access could write 95% of that from training data.
Feed the title and outline to a frontier LLM with no search access. How close is the output? If it is 80% there, the page is commodity.
Stakes
What the author is risking by publishing this.
If your B2B SaaS does under 50K EUR MRR, SEO is usually the wrong channel. I say this knowing it costs me leads, but I would rather be right than be hired.
Every situation is different. It depends on your goals.
What is the author putting at risk? A reputation claim? A refusal of revenue? A number they can be held to? If the author has nothing at stake, the reader has no reason to trust.
The scoring ranges
Section titled “The scoring ranges”- 0-4: Commodity. The page is interchangeable. An LLM could produce a close equivalent. Any competitor could republish it under their own brand without changing substance.
- 5-8: Mostly commodity. This is the most common score range in B2B content, and the hardest to spot from inside the company. The page cites real sources, uses real numbers, and feels rigorous. But the author is summarizing and explaining, not committing. No first-hand stakes. No positions a competitor would refuse to take. The research is sound. The voice is absent.
- 9-11: Differentiated. The page has material a competitor would struggle to reproduce. Specific client situations, named frameworks, working tools, positions that cost something to hold.
- 12-15: Cited. The kind of content other people reference. Built around working tools that run the math the way the firm’s actual specialists run it, with worked examples using real case data and positions competitors would refuse to publish.
The 5-8 trap
Section titled “The 5-8 trap”This range deserves its own section because it is where the majority of professionally produced B2B content lives, and it is the hardest failure mode to diagnose.
A page scoring 5-8 looks competent. It references industry data. It includes statistics. It may even have a named example or two. The problem is that the author is operating as a reporter, not a practitioner. They are explaining what works without evidence that they have done it. They are stating positions without risking anything by holding them.
The 5-8 page passes a quick editorial review. It reads well. It ranks. But it is precisely the kind of content an AI Overview can assemble from multiple sources, because nothing on the page is genuinely the author’s own. The data came from reports. The examples came from case study databases. The viewpoint is consensus opinion stated with confidence.
The fix is not better writing. The fix is better inputs.
The swap test for segment pages
Section titled “The swap test for segment pages”The five-dimension test scores one page. When you build clusters for more than one segment, a second test checks whether the segment pages are really different: would a buyer in this segment find this page more useful than the version written for another segment?
If the answer is “about the same”, the difference is cosmetic. Put the construction page and the healthcare page side by side and read the construction page as a hospital procurement lead. If nothing feels wrong, nothing is targeted either.
What fails the swap test:
- Swapping industry terms in the intro. The most common failure. The first two sentences name a different industry; the body is identical.
- The same content under different segment URLs. The architecture is right, the pages are duplicates with a thin layer of targeting. Search engines notice, and readers notice faster.
- “For [segment]” in the H1 and nothing else changed. The title promises specificity the page doesn’t deliver. That breaks more trust than one undifferentiated page would.
- Tone instead of substance. Shorter sentences and less jargon with the same examples, depth and structure. Tone is not differentiation. Starting point, examples, depth and assumed knowledge are.
Add the swap test to your review checklist, and brief differently, not just write differently: if two segment briefs share the same outline in the same order, the pages will come out the same however good the writer is.
Three operational moves
Section titled “Three operational moves”1. Audit the worst-bleeding page first, not the best-performing one
Section titled “1. Audit the worst-bleeding page first, not the best-performing one”Pages that lost the most clicks while keeping rankings are the highest-leverage rebuilds. They have proven search demand. They have demonstrated ranking ability. The only thing missing is non-commodity material.
Most teams do this backwards: they optimize the page already winning and ignore the page bleeding. Bleeding pages are gold. Google still ranks them, the audience is there, the conversion mechanism has just been removed by the SERP rearranging around them.
Pull your GSC data for the last six months. Sort commercial pages by click decline percentage while filtering for stable or improved rankings. The page at the top of that list is your first rebuild target.
2. Build a sourcing pipeline before writing
Section titled “2. Build a sourcing pipeline before writing”Non-commodity content needs raw material that does not exist in LLM training data. You cannot produce it from desk research, industry reports, and “best practice” summaries. The sourcing has to change before the writing can.
From a client playbook that works: one customer interview per month, with signed permission, specific usage data, and two or three photos. One interview per month fuels a year of non-commodity pages.
Before writing for a segment, I map four context layers. Each has its own source, and each changes how the page gets written:
- Life context: what is happening in this person’s world when they search. New role, post-acquisition integration, quarter-end pressure, hiring freeze, board prep. A VP of Marketing who just started searches differently from one three years into the seat, even with the same query.
- Pressure: the thing that made today the day they searched. A pipeline target they’re behind on, churn they need to explain, a competitor that just raised a round, a CEO question they can’t answer yet. Pressure sets urgency, and urgency decides what kind of page holds attention.
- Prior attempts: which tools, agencies or internal projects already failed or stalled. A page that opens with basics for a reader who has already tried two agencies and a tool gets closed in seven seconds.
- Personal definition of success: what winning looks like for this person, not just the company. Keeping their job, getting promoted, not being the one who picked the wrong vendor. The professional reason gets them to the page. The personal reason gets them to the form.
None of this comes from a brainstorm. It comes from sales call recordings (Fathom, Avoma), CRM deal data, support and onboarding tickets, LinkedIn comments on posts by category leaders, subreddit threads in the niche, win/loss notes, and GSC queries with a CTR below what their position should give. If I can’t point to where the context came from, I’m guessing.
Original research is the other source: a small study, survey, crawl or dataset nobody else has. Mine are the Green Blind Spot index (a monthly carbon crawl) and the audience research behind Unprompted. One dataset feeds many pages, and every page that cites it strengthens the others.
The operational question that makes this real: stop asking “what should we cover?” and start asking “what specific situations have we actually encountered?” Mine recent client work for unusual calls, unexpected outcomes, non-standard solutions. Lead with the situation, not a generalized takeaway.
The organizational lift of capturing that material is bigger than the writing itself. This is the part most content teams skip, then wonder why their non-commodity pages still feel generic. The writing is not the problem. The inputs are.
3. Stop publishing anything that fails the test
Section titled “3. Stop publishing anything that fails the test”Kill commodity content to free up time for strategic rebuilds. One client killed 30+ commodity blog posts in Q1 to free their content lead’s time for two priorities: rebuilding the highest-CPC commercial page and writing two new commercial pages where none existed.
Volume-as-floor was a 2024 strategy. In 2026, every piece must answer: “what does this company know that a competitor does not?” If a piece cannot answer that, it does not enter the pipeline.
You cannot produce non-commodity content with a content calendar, freelance pool, SEO brief template, and monthly publishing target. The pipeline has to change.
Say the same thing everywhere
Section titled “Say the same thing everywhere”Every proprietary number on a page could, technically, be invented. A reader can’t verify it from the page alone, and neither can a model. What makes it credible is that the same claim shows up across your whole marketing surface: LinkedIn posts, the sales deck, talks, case studies, support content. Consistency across channels is how proprietary data gets verified.
Formats help build that consistency. The same claim as a talk, a video, a LinkedIn post and a page is four places to check it, and a person saying it on camera is harder to fake than text. A format isn’t differentiation on its own, though: a commodity point on video is still commodity.
When a page ranks but doesn’t convert
Section titled “When a page ranks but doesn’t convert”I used to teach the standard four-intent model and built keyword research courses on it. It stopped explaining my client work when two pages with the same keyword, intent and volume converted at materially different rates: intent tells you what the query wants, not what the person needs.
People scan pages the way animals forage: they follow information scent, small cues in the title, headings, opening paragraph and proof that tell them the page fits their situation (Pirolli and Card, Xerox PARC). Context is what makes the scent strong. Weak scent shows up in your data as:
- High impressions, low CTR in GSC: the title and description don’t carry enough context for searchers to recognize themselves
- Low scroll depth: the opening paragraph didn’t confirm the scent
- Pogo-sticking within 10 seconds: the page broke the promise of the snippet
- Short dwell time on well-ranked pages: ranking for the word, missing the person
This is why segment-specific pages convert 2 to 3 times better than generic equivalents: the fit between context and content is as tight as it gets. When a page ranks but doesn’t convert, start with the opening paragraph, the examples and the assumed starting point, not a rewrite from scratch.
What it looks like when it works
Section titled “What it looks like when it works”On specific queries tied to named customer stories or first-hand decisions, AI Overviews either cite the page or do not render at all. The content survives because the AIO cannot compress it without losing meaning.
A page scores 14/15 because it is built around a working tool that runs the math the way the firm’s actual specialists run it, with worked examples using their own case data and positions their competitors would refuse to publish. None of that is reproducible by an LLM. None of it would survive a generic agency brief.
The version a content agency would have shipped against the same keyword scores around 4/15. Same length, same headers. No working tool, ranges sourced from “industry research.” Both would have ranked. Only one still gets clicks.
How this connects to ABM SEO
Section titled “How this connects to ABM SEO”The five-dimension test is the quality gate that sits underneath the entire content cluster architecture. You can follow the ABM SEO playbook perfectly, build the right segments, structure the right clusters, wire the right internal links, pass the swap test, and still produce interchangeable content if the inputs are commodity.
Non-commodity content is not a nice-to-have layer on top of ABM SEO. It is the reason the methodology produces results. Segment-organized content gives you the structure. Non-commodity material gives you the substance. Without both, you have a well-organized collection of pages that AI Overviews can still compress.
Score your pages. Run the scoring tool. Rebuild the bleeding ones first.
Get notified when new chapters publish
You're on the SEO lead path. Step 3 of 5.
0 of 5 read
- Next: Content Clusters
- Not read: Internal Linking
- Reading now: The Content Quality Gate
- Not read: The Forecast Spreadsheet
- Not read: Measuring by Segment
Switch path: CMO or Head of Growth