By Sagar Shankaran, Founder of CallSphere
A 1,200-page record in one question: timeline, inconsistencies, exhibit mapping. What it saves on a flat fee, and what still needs an attorney's judgement.
Key takeaways
Every immigration attorney paying attention already ran this experiment. You had a record of proceedings past a thousand pages, fed it to whatever tool you had, and it read the first chunk, forgot it, read the next, and produced a summary of a case that did not exist. It could not hold the file. So you went back to the highlighters and stopped believing the pitch.
That specific limitation ended this year. Claude Opus 4.6 and the models that followed it can hold roughly a three-quarter-million-word document set in mind at once, which in plain terms means an entire record of proceedings — the I-589, every declaration and supplemental declaration, the credible fear interview transcript, the country conditions packet, the exhibit index and the prior counsel's filings — now goes into a single question, with no splitting it into pieces and no arguing about which page you meant. The tool sees page 4 and page 940 at the same moment. That is the only thing that changed, and it is the thing that makes the file useful.
Why it matters here more than almost anywhere else: in removal proceedings the case is very often won or lost on internal consistency, and the person whose job it is to find the inconsistencies is the DHS trial attorney, who will have read the whole thing.
Take an asylum case heading to an individual merits hearing. The file typically holds the Form I-589 as originally filed, sometimes years earlier. The respondent's declaration, and then a supplemental declaration prepared after new facts came out. The credible fear interview record from the port of entry or the detention facility, taken through a phone interpreter, often at four in the morning, transcribed by an asylum officer who was summarising. Medical and psychological evaluations. Police reports and death certificates from the home country, in Spanish or Portuguese or Haitian Creole with certified translations attached. Then the country conditions packet: State Department human rights reports, Human Rights Watch, expert declarations, news articles — three hundred to eight hundred pages, and the index is the only reason anyone can find anything.
Add the procedural layer: the Notice to Appear, master calendar minute orders, motions to continue, a change of venue, a prior attorney's filings that may contain a factual assertion your client now disputes. On a case that has been pending five years, twelve hundred pages is normal and two thousand is not unusual.
An associate preparing this for hearing does one thing above all others: builds a timeline and hunts for every place two documents say something different about the same event. Which month the threat happened. Whether the respondent was detained overnight or for two days. Whether the brother was present. Whether the police were called. These differences are usually innocent — a frightened person recounting trauma through an interpreter at 4 a.m. is not a reliable transcriptionist of their own life — but the inconsistency is what the government cross-examines on, and if your side has not found it first, the hearing is a surprise party.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for law firm in your browser — 60 seconds, no signup.
flowchart TD
A["Whole file loaded at once: I-589, declarations, CFI transcript, country packet, prior filings"] --> B["One question from the attorney"]
B --> C["Timeline of every date the respondent has given, with the source document"]
B --> D["Every place two documents describe the same event differently"]
B --> E["Exhibits with no matching claim in the declaration"]
B --> F["Country report pages that support each specific harm alleged"]
C --> G["Associate builds the cross-examination prep memo"]
D --> G
E --> G
F --> G
The second branch is the one that earns its keep. "Show me every place the declaration, the supplemental declaration and the credible fear interview describe the same event differently, and cite the page for each." What comes back is a table: four discrepancies, three of them explainable in one sentence at direct examination, one of them a genuine problem you now have eleven weeks to address rather than eleven seconds.
The fourth branch matters for a different reason. Immigration judges have grown impatient with country packets that are large and unmoored. Being able to say "the harm alleged at paragraph 14 is corroborated at exhibit F, pages 212 and 219" is worth more than another hundred pages of general reporting, and finding those two page numbers by hand across an eight-hundred-page packet is an afternoon.
The same capability lands on the other half of a mixed practice, in a different shape. A contested dissolution with a tracing issue produces a document pile nobody can hold in their head: eleven years of monthly statements from four accounts, a Schedule of Assets and Debts, both parties' Income and Expense Declarations across three amendments, retirement plan statements, a business valuation report, and the subpoenaed records that arrived as a scanned PDF with no bookmarks.
The question that pile could not answer before is the tracing question. When the down payment came from a premarital account, which deposits during the marriage were community earnings and which were separate, and what is the running balance at each transfer? That is a Moore/Marsden style apportionment in California and a similar exercise elsewhere, and firms have historically either paid a forensic accountant several thousand dollars or waived the issue because the fee did not justify it. Holding all eleven years at once, the assistant produces the deposit-by-deposit ledger with a source page for every line, and the forensic accountant — who you should still hire on the big cases — starts from a draft instead of a banker's box.
Immigration work is largely flat fee, which means every associate hour spent reading is margin gone. Assumptions, stated: an associate loaded at $95 an hour of firm cost, billed or budgeted against a flat fee; nine merits hearings a year that involve a record over 800 pages.
| Per merits-hearing file | Reading it by hand | Whole file in one question |
| Associate hours to build the timeline | 9 | 2 |
| Associate hours on inconsistency hunting | 6 | 2 |
| Associate hours mapping country reports to allegations | 4 | 1.5 |
| Total hours | 19 | 5.5 |
| Firm cost at $95/hour | $1,805 | $523 |
| Cost of running the file through the tool | — | about $15 to $40 |
| Margin recovered per case | roughly $1,250 | |
| Across nine cases a year | about $11,250, plus 121 associate hours | |
The tool cost line is worth dwelling on. Running a twelve-hundred-page file costs tens of dollars, not hundreds, because frontier pricing came down roughly tenfold from 2025. Three years ago this arithmetic did not work on a flat-fee case. It works now.
Three failure modes, and the first is the one that ends careers. Anything the tool produces that looks like a citation to a Board of Immigration Appeals decision or a circuit case must be checked against the actual reporter before it goes near a brief. The rule from Mata v. Avianca in 2023 has not softened; it has hardened, and immigration judges have seen enough fabricated citations to be unforgiving. Use this for reading your own file, not for finding law.
Second, a summary is not the record. If the assistant says the credible fear interview mentions a detention of two days, you open the transcript and read the line before you build direct examination on it. The correct workflow is that it points and you verify, every time, without exception.
Still reading? Stop comparing — try CallSphere live.
See the law firm AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
Second-and-a-half, translation. Certified translations are still required for foreign-language documents, and a tool's rendering of a Guatemalan police report is not a certified translation. It is fine for orientation and useless for the record.
Third, and most important: credibility judgement is the judge's and the strategy is yours. An assistant can show you that two documents disagree. Only an attorney who has sat with the client for six hours knows whether the difference is trauma, interpretation, or a story that is falling apart — and knows what to do about it. Everything in this post makes preparation faster. None of it makes the decision.
It is a question you have to answer deliberately. Use a business or enterprise agreement where your material is not used for training, keep the retention settings tight, and disclose the practice in the engagement letter as ABA Formal Opinion 512 contemplates. For detained clients and asylum matters, be extra careful about anything that could identify family members still in the home country.
Current models read scanned pages, including the ones from the old office copier. Quality still matters — a fax-of-a-fax police report from 2011 will produce a shaky reading, and that is exactly the document you should verify by eye. Bookmark or index the packet if you can; it makes the page citations far more usable.
It can draft sections that are pure record recitation — statement of facts, procedural history, the exhibit-to-allegation map — and those are genuinely useful. The legal argument is yours, and every authority in it gets verified by a human. Treat it as an excellent associate who has read everything and cannot be trusted to cite anything.
This is arguably the strongest use of all, because the briefing clock is short and the transcript is long. Loading the full transcript and asking where the judge's findings depart from the testimony gives you a list of record cites to work from on day one instead of day six.
Take a merits case you already won, load the whole record, and ask for the inconsistency table. Compare it against what you found by hand at the time. If it surfaces something you missed, you have your answer. If it invents something, you learned that cheaply, on a file where nothing is at stake.
One practical side effect: the week a hearing notice goes out, the phone does not stop. CallSphere builds AI voice and chat agents that answer a law firm's phone line and web chat 24/7 in the languages your clients speak, take the message with the case number right, and book the prep appointment — so the associate who is finally getting through a twelve-hundred-page record is not the person answering the third call of the morning.

Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
AI that answers only from a firm's approved memos and the client's file, showing the page — so a wrong travel answer is caught before it costs a decade.
AOS, (c)(9), Epstein credits, MSC receipt prefixes and two-surname names: the shorthand a general model gets wrong, and what industry-tuned AI actually fixes.
Returned mail, unanswered texts and an unsigned G-28 predict a blown hearing weeks out. How nightly docket checks catch the failure before the judge does.
Block billing, clerical time and wrong task codes cost defense firms 6% of every invoice. Reading all 4,200 entries first now costs a few dollars a month.
ERISA reimbursement, Medicare conditional payments and made-whole: how grounding keeps a confidently wrong lien answer out of your disbursement statement.
Tbilisi professional-services firms serving relocating founders and IT companies use CallSphere AI voice and chat agents to answer enquiries 24/7 in English, Georgian and Russian and book consultations.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI