Managing speaker quotes as reusable assets across a portfolio of shows

The request arrives on a Thursday. Marketing wants a good quote about supply chain risk for the registration page, ideally from somebody senior, ideally from one of the shows in the same vertical. Somebody remembers a panel from the spring, finds a PDF of the post-show report, and retypes a sentence out of it.

Speaker quotes as reusable assets is the alternative to that, and the case for it is defect-driven. Retyping introduces wording errors. The PDF has no timecode, so nobody can check the sentence against what was said. The job title in the report was current in the spring and is now wrong. Nobody knows whether the speaker agreed to promotional use, because nobody recorded the answer.

What a quote register has to hold

Nine fields, one row per quote, and every field earns its place by preventing a specific failure.

The exact wording, copied from the corrected transcript rather than retyped. The speaker's name, spelled as they spell it. The employer and role held at the time of speaking. The date and the edition. The session identifier. A start and end timecode into the recording, so the sentence can be heard again in ten seconds. The approval state. Any restriction the speaker attached. And the identifier of the person who logged the row, because you will want to ask them something in eighteen months.

Two of those fields are usually missing in practice, and they are the two that matter most. The timecode makes verification cheap, which is the difference between a register people trust and a register people work around. The employer-at-time-of-speaking field is the one that stops a quote ageing badly.

Where it lives matters less than who is required to write to it. A spreadsheet with nine columns, owned by the editorial coordinator and filled during the transcript pass, beats a purpose-built module that three people have access to and nobody updates in show week. The test is whether a row exists for every quote your desk published last month. If the answer is no, a better tool will not change it.

Populate the register from the transcript pass rather than from published articles. A quote entering the register from an article has already been through an editing stage, so any tightening applied there gets baked in as the canonical wording, and the next person quotes the edit instead of the speaker.

Why job titles are the defect you will actually hit

Trade audiences change employer often, and speakers are drawn disproportionately from the part of the audience that gets promoted. So the register decays in a predictable way, and the useful question is what to do about it.

Work an example. Say your register holds 1,850 quotes across six shows and four editions, and 1,100 of them are older than 24 months. Nobody is going to re-verify 1,100 rows.

So sample. Draw 40 rows at random from the older group and check each speaker's current employer and role against a public source. Say 11 of the 40 have changed, which is 27.5 per cent, implying roughly 300 stale rows across the older group. At four minutes to re-verify and correct a row, cleaning all 300 is about 20 hours of work, and it will be wrong again within a year.

Now cost the alternative. Your desk reuses perhaps 60 archived quotes a year. Verifying at the moment of reuse costs the same four minutes each, so four hours a year, and it catches every stale row that matters at exactly the point it would have caused an error. The sample still has a job to do, because the 27.5 per cent figure is what you show the person who wants to reuse 200 quotes in a campaign without checking any of them.

The rule that falls out: verify at reuse, sample at rest. Attach the staleness estimate to the register as a documented figure with the date it was measured, and re-run the sample annually.

Should a quote be re-approved every time it is reused?

No, and the reason is that approval and permission are answering different questions.

Approval is about wording. Did the speaker confirm the sentence says what they meant. That is settled once, at first publication, and the answer does not expire. The scope of that check is a one-sentence email, described in turning a panel into an article, and its result belongs in the register as a state with a date.

Permission is about use. Reporting a quote in editorial coverage of your own event is ordinary journalism. Putting the same sentence on a registration page under a photograph of the speaker, or in a sponsor's deck, is promotional use, and a speaker who was happy with the first may reasonably object to the second. What the speaker agreed to when they were recorded governs this, and the release wording is speaker release and recording rights territory rather than a register design question.

The register's job is to make the distinction visible in the row. Two flags: approved wording, and cleared for promotional use. A quote with the first and not the second is fully usable in an article and unusable on a landing page, and the person building the landing page can see that without emailing anyone.

Where do machines get your quotes wrong?

Everywhere the attribution is thin, and the scale of the problem is now documented.

Jaźwińska and Chandrasekar published a Tow Center study in the Columbia Journalism Review in March 2025 that ran 1,600 queries across eight generative search tools, giving each tool an excerpt from a real article and asking it to identify the headline, publisher, publication date and URL. Their summary sentence is blunt: "Collectively, they provided incorrect answers to more than 60 percent of queries." The spread was wide, from 37 per cent incorrect for Perplexity to 94 per cent for Grok 3. DeepSeek misattributed the source in 115 of 200 cases, and more than half of the responses from Gemini and Grok 3 cited fabricated or broken URLs.

Read that as a description of what happens to a quote once it leaves your site. A sentence said on your stage, republished by three trade titles and a newsletter, is a string with four candidate origins, and the tools resolving that question get it wrong most of the time. Your show gets no credit for having hosted the moment, and in some fraction of cases the words are attached to the wrong speaker entirely.

Two moves help at the margin. Publish the quote on your own domain before or at the same time as you give it to a trade title, so the earliest indexed instance carries your attribution. And keep one page per session that holds every quote from that session with speaker, employer and date together, so there is a single dense source describing the moment correctly, instead of six partial ones each carrying a different fragment.

You cannot fix the engines. You can make the correct answer easier to find than the incorrect ones, and you can measure how often they get it right, which is measuring citation accuracy honestly as a separate exercise with its own method.

Marking up a quote so the attribution travels with it

The machine-readable half of this is cheap and mostly unused.

Schema.org defines a Quotation type, described in version 30.0 from March 2026 as "a quotation. Often but not necessarily from some written work, attributable to a real world author". It carries spokenByCharacter for the person the quote is attributed to, and inherits isBasedOn, creator, text and isPartOf from CreativeWork. So a pull quote on your article page can state, in structured form, who said it and which work it came from, and the work can point back at the session it was recorded in.

Adoption is low, which cuts both ways. The type is in the newer part of the vocabulary and no search feature depends on it today, so nobody should promise a ranking effect. What it does is make your page the one place where the quote, the speaker, the employer at the time and the source session are stated together in a form a retriever can parse without guessing. That costs a template change.

Do the plain-text version regardless. The sentence around the quote should name the speaker, the employer at the time, the session and the edition, in prose, in the same paragraph. A retriever that ignores your markup still reads the paragraph, and a paragraph that carries its own attribution survives being chunked and quoted out of context, which is how most of your content is now consumed on the content studio side of the business.

Where this stops

A register is only as honest as its wording field, and the wording field has a failure mode that no schema prevents. Every quote in it has been through a filler-removal pass, and a filler-removal pass changes what a sentence sounds like even when it changes nothing about what it says. Storing the timecode alongside the wording is the mitigation, because it lets anyone hear the original in ten seconds. It is a mitigation rather than a fix.

The second limit is that quotes are the easiest asset to over-reuse. A speaker who gave one good sentence in 2024 can end up representing your entire portfolio's view on a subject for three years, appearing in six articles and two campaigns, because the register made them easy to find. The register makes that pattern visible if anyone looks, so add a reuse count to the row and read it occasionally.

Export the quotes your desk published in the last twelve months into a spreadsheet, add a column for the employer each speaker held at the time of speaking, and fill it in from the transcripts. However many rows you cannot fill is the size of the register you do not yet have.