PowerPoint Citations: Source Footers, Links, and AI Verification

A slide citation has one job: let a reviewer find the underlying evidence in under a minute, without asking anyone.

That single criterion resolves most of the arguments about citation style. If a footer does not achieve it, the footer is decoration, and decks are full of decorative citations. "Source: internal analysis" tells a reader nothing. Neither does a company name with no document, no page and no date.

I want to set out what actually works, and then the specific problem of verifying citations in generated slides.

What a source footer needs to carry

Four elements, and the third is the one most often missing.

The source. The publisher, system or document. Specific enough to locate: not "market research" but the named report or the named system.

The location within it. Page, table, exhibit, named range or record identifier. This is what turns a citation from an attribution into a lookup. A forty page report cited without a page number is a reading assignment.

The date of the data. Not the date you accessed it. The period the figure describes, or the version and extraction timestamp for internal data. A market figure with no date is not a data point.

An identifier, for internal records. When the claim comes from your own knowledge base, carrying the record identifier means the reviewer can pull the exact record with its owner and approval status.

A compact format that works:

Source: Eurostat, Industrial production index, Q2 2026, extracted 2026 08 12

Or for internal material:

Source: FY26 reporting pack, Revenue_Actuals, v14, as of 2026 07 31 [REC1042]

Small type, consistent position, part of the master so nobody has to decide where it goes.

Why URLs make poor slide citations

They are the reflex answer and they are usually the wrong one.

They break. Link rot is relentless, and a deck can circulate for years.

They are unusable in the medium. Nobody types a long URL from a projected slide or a printed handout.

They consume space disproportionately and push out the information that would actually help.

They often point at the wrong granularity. A link to a report homepage does not locate the figure on page thirty one.

Better: cite the source and location in readable text on the slide, and keep the resolvable link in the underlying record where it can be maintained and updated centrally. If someone needs the link, they look up the record. If the link changes, you change it once.

Distinguish evidence from argument

A slide carrying an external fact needs a source. A slide carrying your recommendation does not, because you are the source.

The mistake is failing to make the difference visible. When every slide looks the same, a reader cannot tell which statements rest on evidence and which rest on your judgement, and both get treated with the same weight. That is bad for the reader and, when the judgement turns out to be wrong, bad for you.

Make it structural. Evidence slides carry footers. Argument slides do not, and are visibly a different kind of page. A reader should be able to tell at a glance whether they are being shown a fact or a view.

The verification problem in generated slides

Now the part that matters most in 2026.

Generative systems produce citations that look correct. Plausible publisher, plausible report name, plausible page number, positioned exactly where a citation belongs. Sometimes the source does not exist. Sometimes it exists and says something different. Sometimes it exists, says the right thing, and is four years out of date.

This is not an occasional glitch. It is a structural property of how these systems work, and the presentation format amplifies it, because a citation in small type under a chart carries an implicit assurance that somebody checked.

Three checks, for every generated citation:

Does the source exist? Open it. Not a search that returns something similar. The actual source.

Does it contain this specific claim, at this location? A source that discusses the topic is not a source that supports the number.

Is it current? A correct citation to a superseded figure is still wrong on the slide.

An unresolvable citation is a defect, not a formatting issue. It should block release in the same way a missing legal disclaimer would.

Designing the workflow so verification is cheap

Manual verification of thirty citations does not happen under deadline pressure. So build the workflow so most of it is unnecessary.

Retrieve, never compose. The system draws citations from records that already exist and are already approved. It does not construct a reference from what a plausible reference looks like. This single design decision eliminates the entire class of invented citations.

Carry identifiers end to end. The record identifier travels from the knowledge base into the slide footer. Verification becomes a lookup rather than a search.

Mark gaps explicitly. When the system cannot find supporting evidence, it produces a visible marker rather than an unsourced statement or a plausible reference. Unresolved markers block release.

Automate resolution checking. Every identifier in the deck should resolve to a live record. Every external citation should be checkable against a maintained source list. Both are mechanical and neither needs a human.

Surface staleness. When a cited record passes its review date, decks using it should be flagged. This is what stops a 2024 market figure reappearing in a 2027 pitch.

Keeping citations current across reuse

Slides get reused, and this is where citation quality degrades fastest.

A slide built in March with a current figure gets copied in September, then again the following year. The footer still says March, or worse, the footer was updated to look current while the figure was not. Both happen constantly.

The fix is to cite the record rather than the copy. When a slide draws from a maintained record with a review date and an owner, reuse can surface staleness automatically: the system knows the record behind the figure is eighteen months old and can say so before the slide goes into a client deck.

A hard coded footer cannot do that. It is a string, and strings do not know when they stopped being true.

Getting the visual design right

Practical notes, because a citation nobody can read is not a citation.

Consistent position, defined in the master. Bottom left or bottom right, same place on every slide. Nobody should decide this per slide.

Small but legible. Small enough not to compete with the content, large enough to read when projected. Test it on a screen at the back of a room rather than on a laptop.

Sufficient contrast. Light grey on white is a common choice and frequently fails accessibility contrast requirements. Citations are content, not decoration.

Real text, never an image. Assistive technology needs to read it, and so does anyone searching the deck.

Room for a long source. Some citations are long. The layout should handle it without overlapping content or silently truncating, which is a common and quietly damaging failure.

The citation checklist

Where offgen fits

Citations in offgen come from records rather than from generation. When a claim is drawn from the Company Brain, the record identifier travels into the slide footer, so a reviewer can pull the exact record with its owner, approval status and review date. When no supporting record exists, the system marks a gap rather than composing a reference that looks right.

Brand governance handles footer position, size and contrast as template rules, so citation formatting is a property of the deck rather than a decision someone makes at the end. Output stays natively editable, so footers are real text that can be read, searched and corrected.

More on how records carry provenance is on the CVs and case references page.

The habit I would build into any team working with generated material: treat an unresolvable citation exactly as you would treat a missing legal disclaimer. Not a tidy up item. A blocker. Once that is the norm, the rest of the discipline follows on its own.