Skip to main content
Skip to main content
All posts

Inside Cognos: optimizing for answer engines that never publish a changelog

ChatGPT, Perplexity, Copilot and Claude all decide who gets cited, and none of them announce when the rules change. Here is how the Cognos program is built to keep up, and what it does each month.

Andrew Bethel avatarAndrew BethelCo-Founder & COOAug 4, 20267 min read
Inside Cognos: optimizing for answer engines that never publish a changelog, cover image

Most of what gets written about AI search is really about Google. That's a reporting bias, not a market reality. Google documents its systems, so Google is what commentators can cite.

Your buyers aren't that tidy. They ask ChatGPT for a shortlist. They ask Perplexity for a comparison. They ask Copilot inside the tools they already work in, and Claude when they want the reasoning shown. Each of those systems retrieves, ranks and cites differently, and none of them ships release notes telling you what changed.

That's the problem Cognos was built around.

The structural problem: no changelog#

Classic SEO had a rhythm you could plan against. Google confirmed a handful of core updates a year, the industry braced, and agencies scheduled work around them.

Answer engines don't work that way. Retrieval behavior shifts as models are retrained, as index partners change, as grounding and citation policies are tuned. There's no announcement. The first signal most companies get is a quarter of soft pipeline.

So the only durable strategy isn't "optimize correctly once." It is measure what the engines are actually doing, continuously, and adjust as they move.

"Nobody tells you retrieval changed. You either measure it weekly or you hear it from your pipeline a quarter later."

Andrew Bethel, COO, Perfectus Labs

What Cognos actually is#

Cognos isn't a dashboard you log into. It's a continuous program our team runs on your site, with four parts working together.

1. Technical resolution, not a technical report#

Cognos audits the factors that govern whether a machine can read and trust your site (crawlability, rendering, indexation, schema integrity, canonical structure) and fixes what it finds directly. The output is the fix, not a PDF recommending one.

This matters more for AEO than for SEO. Traditional crawlers are forgiving; they read the page and fall back to metadata. AI retrieval leans on structure first. A site can rank respectably in classic search and still be effectively unreadable to the systems doing the answering.

2. The Fan-Out Map#

Nobody searches an answer engine the way they searched a search box. One buying question fans out into a hundred or more sub-questions: comparisons, objections, pricing framing, "is X worth it for a company like mine."

The Cognos Fan-Out Engine maps that tree for your category: the prompts, the follow-ups, and the branches where a competitor is currently supplying the answer, or nobody is. That white space is the plan. Content is written against retrieval demand, not against a keyword list.

3. Reinforcement across the channels engines actually read#

Being right on your own website isn't sufficient. Retrieval systems corroborate. So coverage runs across the property set they pull from (your site, Google Business Profile, press, social and authority channels), so a claim about your business is consistent everywhere a model might check it.

Consistency is the underrated variable here. Conflicting facts across sources don't average out; they reduce the confidence a model has in citing you at all.

4. Adaptation as the default state#

The engines change. Cognos re-measures and re-optimizes on that cadence instead of on a contract cadence. Retrieval coverage that's maintained compounds. Retrieval coverage that ships once decays.

Why this is hard to run in-house#

Three reasons, consistently:

  • Cadence. A manual optimization round is usually stale before it ships.
  • Mapping depth. Building a real fan-out map, hundreds of sub-queries per service captured from live retrieval behavior, is weeks of specialized work per category.
  • Channel breadth. Reinforcing priority queries across site, profile, press and authority sources at the same time is a staffing problem, not a tooling problem.

Any one of those is manageable. All three, permanently, is a team.

What it looks like on your side#

Light. The main ask is reviewing content before publication for factual accuracy and brand voice. Strategy, technical work, production, profile management, distribution and reporting sit with our team. There's no platform for you to learn.

The honest framing#

Nobody can promise placement inside a system they don't control. What is controllable is whether your business is technically readable, structurally trustworthy, comprehensively answered across the questions buyers actually ask, and consistent everywhere a model looks.

That's the entire job. Cognos is the program that does it continuously instead of once.

Want to see what the engines currently say about your category? Request an AEO audit and we'll show you the fan-out map for your market.

About the author

Andrew Bethel, Co-Founder & COO
Andrew Bethel

Co-Founder & COO, Perfectus Labs · Dayton, Ohio · remote-first team

Andrew is COO of Perfectus Labs, where he runs the operating system behind Cognos AEO and the research shaping what the company builds next.

Connect on LinkedIn

Experience behind this post

  • Runs company-wide operations across Cognos AEO, delivery, team development, and strategic planning.
  • Designed and implemented the operating playbooks for AI-powered service and software lines, then scaled the teams that run them.
  • Connects live client and product feedback to the research agenda for the next generation of systems, tools, and AI workflows.

Writes reliably on

How AI companies turn research into dependable products and client deliveryOperating Cognos AEO as an ongoing program instead of a one-off auditHow leadership teams evaluate and apply the next generation of AI toolsMeasuring AI search visibility and connecting it to business outcomes

Published under these standards

  • Every claim about a platform change links to the primary source — the vendor's own documentation, changelog, or policy page.
  • Numbers come from engagements he works on or from published research that is cited by name and date.
  • Posts distinguish between what the team has observed in practice, what is documented publicly, and what remains a hypothesis.