The Screenshot Is in English.
The Conversation May Not Be.
July 2026 · VdVLaw
AI, juvenile messages, and the legal danger of misunderstanding digital evidence
“The limits of my language mean the limits of my world.”
- Ludwig Wittgenstein, Tractatus Logico-Philosophicus
A screenshot can look simple.
There is the message. There is the date. There is the username. There is the little bubble. Everyone nods as if the conversation has now politely explained itself.
It has not.
Sometimes the screenshot is in English, but the conversation is not. .....Not really.
When juveniles are messaging on Instagram, texting, using slang, abbreviations, emojis, inside jokes, sarcasm, half-sentences, missing context, and group-chat logic, the words may technically be English. But the meaning can feel like a foreign language spoken at 2:00 a.m. by people who refuse punctuation on principle.
I have a 16-year-old son, so I get a live demonstration (Example Link) of this almost daily. I regularly have to remind him that I am happy to be addressed in English, or Dutch if he wants the premium family-language option, but I still need to understand what the hell he is talking about. That is not a complaint about teenagers. It is a reminder that shared words are not always shared meaning.
That matters in legal work.
Because a digital conversation is not just what appears in the screenshot. It is sequence, context, speaker identity, timing, missing messages, slang, tone, prior relationship, and what the people actually meant when they communicated.
A pile of screenshots is not a conversation.
.....It is a puzzle pretending to be evidence.
Screenshots are not self-explanatory
The problem with screenshots is that they often arrive looking complete.
They are not.
They may be cropped. They may be out of order. They may omit prior messages. They may show only one side of the conversation. They may preserve the dramatic part but not the boring context that explains it. They may include slang or coded language that makes sense to the participants but not to the adults now trying to reconstruct it.
And in juvenile cases, that gap can matter.
A phrase that sounds threatening may be sarcasm. A phrase that sounds harmless may be a reference to something outside the screenshot. An emoji may change tone. A missing message may change meaning. A nickname may identify a speaker. A screenshot taken at the wrong point in the sequence may turn a response into what looks like an opening statement.
This is where digital evidence gets dangerous. Not because screenshots are useless, but because they are easy to overread.
The screenshot may show words. It may not show meaning.
“The screenshot is in English. The conversation may not be.”
California evidence law recognizes that writings need authentication. Evidence Code section 1400 defines authentication as evidence sufficient to sustain a finding that the writing is what the proponent claims it is, and Evidence Code section 1401 requires authentication before a writing may be received in evidence. California Courts social-media evidence materials make the same basic point in a practical format: social-media evidence still has to be connected to what the proponent says it is.
Cases applying those rules are a reminder that digital content is not admitted simply because it appears on a screen. In People v. Valdez, the Court of Appeal considered authentication of MySpace evidence using the surrounding circumstances connecting the page and content to the defendant. In People v. Goldsmith, the California Supreme Court declined to create a special heightened authentication rule for automated digital photographs merely because digital evidence could theoretically be manipulated, but the court still focused on foundation and reliability.
That distinction is useful. The point is not to panic every time someone sends a screenshot. The point is to stop pretending the screenshot has done all the work for us.
The legal-support version comes earlier: before the hearing, before the motion, before the witness questions, before anyone decides what the screenshots “prove.”
Someone has to organize them. Someone has to sequence them. Someone has to ask what is missing. Someone has to translate the teenager, which is not the same thing as translating the words.
That sounds like a joke right up until a juvenile adjudication turns on it.
Where AI can actually help
This is one place where AI can be useful.
Not decisive.
Useful.
AI can assist with sequencing screenshots, separating speakers, identifying repeated messages, suggesting possible meanings of slang or abbreviations, flagging unclear references, creating a working timeline, and turning a pile of images into a conversation map.
That matters because “87 screenshots in no obvious order” is not a legal exhibit. It is a cry for help with a file name.
AI can help turn that chaos into something more usable:
-
Who is speaking?
-
When did the conversation happen?
-
Are there duplicate screenshots?
-
Are there gaps?
-
Is the order clear?
-
Are there missing timestamps?
-
Are any messages responses to something not shown?
-
Does the slang have multiple possible meanings?
-
Does the explanation need to be marked as uncertain?
But that is only the beginning.
Someone still has to compare the AI output to the actual screenshots. Someone still has to verify the sequence. Someone still has to mark uncertainty as uncertainty. Someone still has to understand that a slang interpretation is not a finding of fact. Someone still has to ask whether the attorney needs a timeline, an exhibit chart, an authentication checklist, or questions for an investigator.
AI can help read.
It cannot be the only reader.
“A pile of screenshots is not a conversation. It is a puzzle pretending to be evidence.”
That distinction is becoming more important because courts and lawyers are now dealing with a broader digital-evidence problem. AI tools make it easier to create, alter, or enhance digital content, and the National Center for State Courts has published judge-facing guidance on acknowledged and unacknowledged AI-generated evidence, including questions for evaluating authenticity and reliability.
That does not mean every screenshot is fake. It means the old casual confidence around digital evidence is not good enough anymore.
The source matters. The original file matters. The sequence matters. The context matters. The chain of custody matters. And when the speakers are juveniles, the language itself may need careful interpretation.
AI-assisted, not AI-decided
The phrase that matters is AI-assisted. Not AI-decided. Not AI-certified.
Not AI-waved-over-the-evidence-like-a-spiritual-cleansing-ritual.
Assisted.
AI can help with large-volume legal work because so much of legal support now involves pattern recognition, sequencing, comparison, and organization. It can help create indexes, identify repeated names, compare versions, summarize long documents, organize screenshots, and flag inconsistencies.
But the value is not the summary.
The value is knowing whether the summary is useful.
“The life of the law has not been logic: it has been experience.”
- Oliver Wendell Holmes Jr., The Common Law
Holmes was not warning us about AI. He was reminding lawyers that law is not just clean logic. It is experience, context, and human judgment. That is exactly where screenshot evidence can go wrong. The logic of a message may look simple. The experience of the conversation may not be.
That means asking:
-
Does the source support the conclusion?
-
Is the record complete?
-
Is there a missing attachment?
-
Are the screenshots in order?
-
Is the translation literal or contextual?
-
Is the slang being understood correctly?
-
Does the timeline work?
-
Was the report AI-assisted?
-
Are there draft versions or audit logs?
-
Does the attorney need this as a chronology, exhibit list, witness-question outline, or motion-supporting record map?
That is not AI magic. That is judgment.
“I use AI the way a careful person uses a flashlight.
It helps me see faster. It does not decide what I am looking at.”
The ABA made the professional-responsibility point clearly in Formal Opinion 512: lawyers using generative AI must consider duties involving competence, confidentiality, communication, supervision, candor to the tribunal, and fees. In plain English, AI does not absorb the lawyer’s responsibility. The human still has to understand the tool, supervise the work, protect the client, and verify the result.
That is why careful AI use can help legal work, and careless AI use can damage it.
The tool is not the problem.
The unverified confidence is the problem.
When the source itself has an AI backstory
The next layer is even more important.
Sometimes AI is not just helping review the evidence. ...Sometimes AI may have helped create the evidence.
That is already happening with police reports. California Penal Code section 13663, effective January 1, 2026, now addresses official law-enforcement reports generated using AI either fully or partially. It requires disclosure of the AI program used and a prominent statement that the report was written fully or partly using AI, requires the preparer’s verifying signature, requires retention of the first AI-created draft for as long as the official report is retained, and requires an audit trail identifying the person who used AI and the video or audio footage used to create the report, if any.
That is a major development.
Because if a police report was AI-assisted, the final report may not be the whole record.
A defense attorney may need to know:
-
What did the body-worn camera show?
-
What did the audio transcript say?
-
What did the first AI draft say?
-
What did the officer change?
-
What did the officer leave in?
-
Was the AI-generated draft disclosed?
-
Were there audit logs?
-
Was the final report consistent with the source footage?
-
Were words like “aggressive,” “combative,” “furtive,” or “resisting” supported by the video?
That is not an abstract technology issue.
That is discovery.
The report is not the footage. The summary is not the record. The conclusion is not the proof.
And if the first AI draft, audit trail, or source-footage history shows that the final report changed language in a meaningful way, omitted context, added unsupported descriptions, relied on the wrong footage, or revealed something useful for impeachment or defense investigation, that may move beyond good practice and into Brady/Giglio territory. The point is not to call every AI draft exculpatory. The point is to stop pretending the final report is the only document that matters when the law now requires retention of the draft and audit trail behind it.
That is the careful distinction. Brady requires disclosure of favorable material evidence. Giglio applies the same due-process concern to impeachment evidence. Kyles makes clear that the prosecutor’s duty includes favorable evidence known to police investigators and others acting on the government’s behalf.
California criminal discovery also matters. Penal Code section 1054.1 requires the prosecuting attorney to disclose specified materials and information if they are in the prosecutor’s possession or if the prosecutor knows they are in the possession of the investigating agencies. Penal Code section 1054.5 provides the mechanism to enforce discovery after the required informal discovery process.
So the practical point is simple: the AI draft, audit trail, and source-footage link may not automatically be Brady material in every case. But if they reveal omissions, inconsistencies, unsupported report language, impeachment material, or facts favorable to the defense, they may become exactly the kind of material the defense needs to ask for and the prosecution needs to disclose.
At minimum, they are now identifiable records. That matters.
If the source was AI-assisted, verification has to start before the summary.
This connects directly to the screenshot problem. In both situations, the danger is the same: people treat the finished-looking thing as if it explains itself.
The screenshot looks like the conversation. The report looks like the event. The summary looks like the record.
But legal work lives in the gap between what something appears to show and what the source actually supports.
The record behind the conclusion
The same principle applies in habeas work.
A government response may reach a conclusion. A prior proceeding may be described a certain way. A claim may be dismissed as unsupported, contradicted, forfeited, harmless, or already resolved.
Fine.
But what record supports that conclusion?
“A conclusion is not a record. It is an invitation to ask what record supports it.”
In California habeas proceedings, Rule of Court 4.551(b)(1)(B) may be relevant because it allows the court, before ruling on a petition, to request an informal response from the custodian of any record pertaining to the petitioner’s case and direct that custodian to produce the record or a certified copy. If records are provided to the court, copies must also be furnished to the petitioner.
That does not turn the rule into a magic discovery wand. Courts are famously skeptical of magic, unless it appears in a properly paginated appendix.
But it does reinforce the practical point: if a conclusion depends on records, the records matter.
The useful work is not just yelling “the government is wrong” into the legal void, satisfying as that may occasionally feel. The useful work is mapping the conclusion to the claimed record, identifying what is missing, and helping the attorney see whether the record actually supports the position being taken.
AI can help build that map. It can help extract assertions from a response, list claimed record references, organize dates, compare procedural history, and flag missing exhibits.
But AI cannot know by itself whether the missing record matters legally.
Someone still has to understand the posture, the claim, the record, and the attorney’s theory.
AI can help find the loose thread.
Why this matters for attorneys
Attorneys do not need more words.
They need usable work product.
They need the person reviewing the material to understand the difference between a document and a conclusion, between a screenshot and a conversation, between a translation and meaning, between a summary and the source, between a police report and the footage behind it.
That is where VDVLaw fits.
VDVLaw’s criminal discovery review work is built around making messy evidence usable: indexed summaries, Bates references, timelines, flagged contradictions, missing records, photo and video issues, and attorney-ready work product. The goal is not to replace the attorney’s judgment. The goal is to give the attorney a cleaner, sharper record to use when making strategic decisions.
“Reading maketh a full man; conference a ready man; and writing an exact man.”
- Francis Bacon, Of Studies
Bacon was writing about education, not discovery review. But the sequence still works. Read the source. Talk through the problem. Write it accurately.
Legal support often lives in that unglamorous middle ground between a chaotic record and an attorney who needs something reliable enough to use. The work is not just making the material look organized. It is making sure the organized material is true to the source.
That matters whether the material is a police report, body-worn camera footage, jail calls, medical records, social-media screenshots, translated conversations, or a habeas response built on conclusions that need to be tested against the underlying record.
The attorney still decides strategy. The attorney still argues the law. The attorney still owns the case.
But the attorney should not have to personally untangle every screenshot, every mislabeled file, every untranslated slang exchange, every missing Bates range, every report that may have an AI draft history, or every government conclusion that quietly depends on a record nobody has produced.
“A clean folder is nice. A verified fact that changes the case is better.”
The real use of AI in legal support
AI is not the product.
The work product is the product.
AI can help produce better work product when it is used carefully, privately, and with human review. It can make a good reviewer faster. It can help organize chaos. It can help detect patterns. It can help make a confusing record legible.
But AI cannot replace the person who knows what to ask.
It cannot replace the person who sees that a conclusion has no record behind it.
It cannot replace the person who notices that screenshots are out of order.
It cannot replace the person who understands that a juvenile conversation may not mean what it appears to mean to adults.
It cannot replace the person who checks whether an AI-assisted police report was compared to the bodycam.
It cannot replace the person who says:
“Wait. Where is the source for that?”
That may be the most important sentence in modern legal support.
Conclusion
The future of legal support is not pretending AI does not exist.
It is using AI carefully while becoming even more serious about source review, confidentiality, context, and human judgment.
Because the more summaries we have, the more the source matters.
The more digital evidence we receive, the more organization matters.
The more AI appears in police reports, legal research, translations, and document review, the more verification matters.
A conclusion is not a record.
A screenshot is not a conversation.
A summary is not the source.
And AI is not judgment.
It is a tool. Sometimes a useful one. Sometimes a very confident one. Occasionally a ridiculous one.
The judgment still belongs to the person willing to check the record.
Sources
[1] California Evidence Code section 1400, authentication defined. Official California Legislative Information. https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=EVID§ionNum=1400
[2] California Evidence Code section 1401, authentication required before admission. Official California Legislative Information. https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=EVID§ionNum=1401
[3] California Courts, Social Media Evidence in Child Support Cases handout, summarizing Evidence Code sections 1400 and 1401 for social-media evidence. https://courts.ca.gov/sites/default/files/courts/default/2024-08/social_media_handout2.pdf
[4] People v. Valdez (2011) 201 Cal.App.4th 1429, authentication of MySpace/social-media evidence based on surrounding circumstances. https://courts.ca.gov/sites/default/files/courts/default/2024-12/btb_23_3d_3.pdf
[5] People v. Goldsmith (2014) 59 Cal.4th 258, California Supreme Court discussion of digital/automated photographic evidence authentication. https://law.justia.com/cases/california/supreme-court/2014/s201443.html
[6] National Center for State Courts, AI-generated evidence: A guide for judges. https://www.ncsc.org/resources-courts/ai-generated-evidence-guide-judges
[7] National Center for State Courts, Bench Card: Evaluating Unacknowledged AI-Generated Evidence. https://ncsc.contentdm.oclc.org/digital/api/collection/tech/id/1244/download
[8] National Center for State Courts, Bench Card: Evaluating Acknowledged AI-Generated Evidence. https://ncsc.contentdm.oclc.org/digital/api/collection/tech/id/1243/download
[9] American Bar Association Formal Opinion 512 / ABA news release on lawyers’ duties when using generative AI. https://www.americanbar.org/news/abanews/aba-news-archives/2024/07/aba-issues-first-ethics-guidance-ai-tools/
[10] California Penal Code section 13663, AI-assisted official law-enforcement reports: disclosure, verifying signature, first-draft retention, and audit trail. https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=PEN§ionNum=13663
[11] California Rule of Court 4.551(b)(1)(B), habeas corpus informal response and production by custodian of records. https://courts.ca.gov/cms/rules/index/four/rule4_551
[12] VDVLaw, Criminal Discovery Review and Indexed Summaries service page. https://www.vdvlaw.net/discovery-review
[13] VDVLaw, What Is Criminal Discovery Review and Why Does It Matter? https://www.vdvlaw.net/criminal-discovery-review
[14] VDVLaw, The police report may now be AI-generated. https://www.vdvlaw.net/ai-police-reports-body-camera-footage-defense-review
[15] Brady v. Maryland (1963) 373 U.S. 83, suppression of favorable material evidence and due process. https://supreme.justia.com/cases/federal/us/373/83/
[16] Giglio v. United States (1972) 405 U.S. 150, impeachment evidence under the Brady framework. https://supreme.justia.com/cases/federal/us/405/150/
[17] Kyles v. Whitley (1995) 514 U.S. 419, prosecutor’s duty to learn of favorable evidence known to police and others acting on the government’s behalf. https://supreme.justia.com/cases/federal/us/514/419/
[18] California Penal Code section 1054.1, prosecution disclosure obligations, including material in the possession of investigating agencies known to the prosecutor. https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=PEN§ionNum=1054.1
[19] California Penal Code section 1054.5, enforcement of criminal discovery obligations after informal discovery procedure. https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=PEN§ionNum=1054.5
[20] Ludwig Wittgenstein, Tractatus Logico-Philosophicus, proposition 5.6, quote: “The limits of my language mean the limits of my world.” https://www.gutenberg.org/ebooks/5740
[21] Oliver Wendell Holmes Jr., The Common Law, quote: “The life of the law has not been logic: it has been experience.” https://www.gutenberg.org/files/2449/2449-h/2449-h.htm
[22] Francis Bacon, Of Studies, quote: “Reading maketh a full man; conference a ready man; and writing an exact man.” https://en.wikisource.org/wiki/The_Essays_of_Francis_Bacon/L_Of_Studies
[23] Polyglot Gives an Entire School Speech in Gen Alpha Slang (Students Lose It)
VdVLaw provides attorney-directed legal support services and is not a law firm. Robert van der Vijver is not an attorney. Nothing in this article is legal advice, and nothing here creates an attorney-client relationship. This article is general information written for attorneys and does not address the facts of any particular case. Criminal discovery obligations, public records access, evidentiary issues, retention issues, ethical duties, and law enforcement technology policies can vary by case, agency, county, court, and date. Statutes, rules, policies, AI product terms, and technology features change over time. Attorneys remain solely responsible for their own legal analysis, discovery decisions, ethical duties, litigation strategy, and verification of all statutes, rules, agency policies, discovery responses, and source materials.