Corpus debt is the accumulated inconsistency across everything a company has published: years of posts describing the company and what it sells in conflicting ways, plus claims that quietly expired. It makes AI systems less certain what the company is, and less willing to cite it.
What is corpus debt?
Quick definition: Corpus debt is the site-wide inconsistency that builds up when years of content describe a company, its category, and its products in conflicting ways, alongside expired claims and outdated positioning. Because AI search systems retrieve and quote individual pages, that inconsistency weakens a domain’s eligibility to be cited as the authority on its own subject.
Corpus debt is an MSMC term. Michael Semer laid out the problem in his September 29, 2026 post on AI search and content debt, which argues that the real exposure in a large back catalog is the story the whole corpus tells, more than any single stale page.
His example is a software company whose category language moved four times. In 2017 it called itself marketing automation. In 2020 the language shifted to customer experience. In 2023 it became revenue orchestration. Today it wants AI systems to describe it as an agentic customer engagement platform. Every one of those eras is still sitting on the blog. A human reader sees a company that rebranded. A crawler, in Semer’s phrasing, sees four different companies that happen to share a domain.
Why corpus debt matters
The cost used to be cosmetic. An outdated post made the site look neglected, and maybe it ranked worse.
AI search changes the stakes because answers are assembled from retrieved pages, and the engine doesn’t know which era of your positioning is current. A buyer asks an AI engine what your company does and gets your 2020 language back, attributed to you. A prospect’s CFO asks about your SOC 2 status and the answer quotes a 2019 post saying certification is on the roadmap.
For a B2B SaaS company in the middle of a category play, this is a positioning failure that marketing can’t see in its own analytics. The new homepage says one thing. The 400 posts underneath it say three others, and the engine has no particular reason to trust the homepage over all of them. Entity optimization efforts stall for the same reason: you can’t define an entity cleanly while your own archive keeps redefining it.
How corpus debt works
The mechanism runs through retrieval, and it compounds:
- An AI engine receives a question and, through query fan-out, turns it into several sub-queries. Google’s documentation says both AI Overviews and AI Mode may use the technique.
- Retrieval-augmented generation pulls whichever passages best match each sub-query. Retrieval matches on relevance to the question. It has no reliable sense of which page reflects current positioning.
- A 2018 post that matches a sub-query precisely can be the one that gets quoted, along with its old category name, retired product tier, and expired statistic.
- Across many questions, the engine encounters the company described in several incompatible ways. Weak, contradictory evidence about what a company is makes it a less attractive source to cite on its own category.
Old content has been testifying the whole time.
Corpus debt vs. content debt
Content debt
Content debt is the older and more generic idea, borrowed from technical debt: the maintenance backlog created by publishing faster than you maintain. Outdated posts, zombie landing pages, broken links, and the thought leadership piece from 2018 that nobody owns. It’s measured page by page, and the usual remedy is a refresh calendar.
Corpus debt
Corpus debt is measured across the whole body of content. Each page can be well written, accurate for its moment, and correctly indexed, and the corpus can still carry heavy debt because the pages disagree with each other about the company. A refresh calendar doesn’t touch it, since refreshing a post in isolation can introduce a fifth category name.
Content debt asks whether a page is out of date. Corpus debt asks whether the site, read as a whole by a machine, describes one company.
How to pay down corpus debt
The repayment instrument is a corpus audit. The sequence:
- Inventory the corpus: every post, landing page, resource, and PDF still indexed, including the ones nobody remembers publishing.
- Define the concepts you want to own, such as your category name, the problem you solve, and the two or three terms you want AI systems to associate with you, and write the canonical definition of each.
- Find the contradictions: competing category definitions, outdated product descriptions, conflicting statistics, obsolete positioning, and clusters of near-duplicate posts competing for the same sub-query.
- Decide each page’s fate. Refresh what still earns traffic or links, consolidate near-duplicates into one strong page, redirect what has a better successor, and retire what no longer describes the company.
Retiring is the step teams resist, because deleting content feels like deleting value. Usually it’s deleting liability.
MSMC ran this on itself first. In September 2026 we audited michaelsemer.com, moved 176 of our own blog posts to draft, and created 55 redirects. Some of those posts were fine. They just described a version of the firm that no longer exists.
How MSMC approaches corpus debt
Corpus debt shows up first in the AI Visibility Audit, usually in two of its seven levers. Entity Definition scores whether AI engines can say clearly what the company is and does. Freshness and Decay scores how much of the content they retrieve is current. A company with a sharp new positioning statement and a low score on both has corpus debt, whether or not it has a name for it.
The Content Effectiveness Audit is where it gets paid down. Its portfolio-level tier evaluates the corpus as a whole, including whether the content tells one consistent story, and ends in a single highest-impact finding. Our view: for most Series B and later SaaS companies, the most valuable content decision isn’t about the next post. It’s about what to do with the last few hundred.
FAQ
What is corpus debt?
Corpus debt is the site-wide inconsistency that accumulates when years of content describe a company, its category, and its products in conflicting ways, along with expired claims. It matters because AI search systems retrieve and quote individual pages, so an outdated or contradictory page can shape how an engine describes the company and whether it cites the domain at all.
How is corpus debt different from content debt?
Content debt is the page-level maintenance backlog: outdated posts, zombie landing pages, and stale resources. Corpus debt is a property of the whole body of content. Every individual page can be accurate for its time and the corpus can still carry debt, because the pages disagree with each other about what the company is and does.
Why would an AI engine quote an outdated page?
Retrieval selects passages by how well they match the question or sub-query, and it has no dependable way to know which page reflects current positioning. If a 2018 post answers a narrow sub-question more directly than anything newer, it can be the page that gets retrieved and quoted, complete with the old category name or an expired claim.
Should we delete old blog posts to fix corpus debt?
Some of them. A corpus audit sorts every page into one of four outcomes: refresh, consolidate, redirect, or retire. Posts that still earn traffic or links usually get refreshed or merged. Posts that describe an old version of the company with nothing worth keeping get retired, with a redirect where a better successor page exists.
How long does a corpus audit take?
It depends on the size of the archive and how many category shifts the company has been through. The inventory and contradiction mapping are the slow parts. The definitions step goes faster when positioning is already settled, and much slower when the audit exposes that the team never agreed on the current category name in the first place.
Key takeaways
- Corpus debt is a whole-site problem: pages that each look fine can still disagree about what the company is.
- AI retrieval quotes whichever page matches the sub-query, so an old post can speak for the company today.
- A corpus audit pays it down by inventorying content, defining the concepts you want to own, mapping the contradictions, and then refreshing, consolidating, redirecting, or retiring.
- Retiring content is usually removing liability.
Browse more definitions in the MSMC glossary.
Understanding this term is the starting point. To find out whether AI systems actually recognize and cite your company on this topic, run an AI visibility audit.