01What happens.
A PDF displays text by mapping characters to a font. Where the font is not included in the file, the mapping can fail, and the characters a machine extracts are not the characters a person sees.
The page looks right and reads wrong, and the loss lands on the headings, because that is where the brand's font is used.
02Why it matters.
If you own the annual report, the typeface that defined the look of the report was never embedded in the file, and that omission removed the report's structure for the machine. Nothing on screen shows the loss.
The reader who finds it is the AI answering an investor's question, and it does not report that the headings are missing; it reports the figures without them.
The same file, ingested into an investor's system, a data vendor's pipeline or the company's own AI, carries the fault in. The data is disordered on reading, and disordered again on each move, and the systems it lands in have no way of knowing.
03The case.
AAAnow read FTSE 100 annual reports as AI reads them. In 11% of reviewed cases a font had not been embedded. The finding is real and the companies are not identified.
The report in question had been designed to the brand guidelines. The headline typeface, licensed to the company, carried the section titles, the chapter openers and the labels on the financial tables. Body text sat in a standard font. The print file and the web file were produced from the same source, and both looked correct on each screen they were opened on.
Read by a machine, the headline typeface was not there to read. The characters mapped to the wrong values.
A section title came through as a string of symbols, a table label came through as nothing the machine could use, and the report's own structure, which tells a reader which figures belong to which section, was gone. AI, asked about the company's revenue by division, found the figures and could not find what they were.
It answered from the body text where it could and from the bare figures where it could not, and where the two disagreed it chose the figures.
The investor who asked received an answer with numbers in it and no way of knowing that the labels had never reached the machine.
The company secretary who had signed off the file had opened it a dozen times. The reporting team had checked each figure against the accounts. The design agency had checked the rendering on 3 devices.
None of those checks reads the file the way a machine does, and the fault cannot be seen any other way.
This is a single, specific fault, found by one check inside the machine's reading of the file. It is distinct from the flattened table and from the oversized file, and it can sit in a report that has neither. A company can pass the other 2 and fail this one, and the report will still look right to the people who made it.
04Exposure.
The figures AI reports without their headings are attributed to the company, and the company answers for the answer. An investor who receives divisional figures with the divisions missing, and acts on them, has acted on the company's disclosure as the machine held it. The board's accountability for the report does not depend on whether the fault was visible.
The cost is correction under the company's name, and the time of the people who have to establish what the file actually contained. Where the report has been ingested into other systems, the correction has to reach each of them, and the company has no list of where the file has gone.
05What AI Readiness changes.
From the inside, the characters on the screen and the characters in the file look the same, and the company has no way of telling them apart. AAAnow reads the report as AI reads it and evidences where a font has not been embedded and what the machine extracts in its place. With that evidence the company can correct the file before publication, keep the brand's typeface and embed it, and check each subsequent report before it goes out.
Your team makes the changes; AAAnow provides an assessment, the evidence, the guidance and the review.
06Related use cases.
The annual report AI read was not the one the board signed
The board-level consequence this fault feeds.
The results table AI read had its figures in the wrong columns
The same report, a different fault.
AI read half the annual report and made up the rest
The third fault in the same family.
07References.
- 11% of reviewed cases with a non-embedded font: AAAnow, FTSE 100 annual report review, 2026, anonymised.
Next step
Talk it through, in your terms.
The briefing sets this case in your organisation’s terms, and the discussion follows your questions.
Observed, FTSE 100 findings, anonymised. No company is identified.