Latest / AI Fire Daily / #460 Neil: Gemini Prompts Powering Free Apps And Branded Slides
Transcript
- 0:00Picture this. We build these massive, incomprehensible
- 0:03AI brains, hundreds of billions of parameters,
- 0:08just mind -bending computing power that has literally
- 0:11read the entirety of human knowledge. Right.
- 0:13And yet, what do most of us actually use them
- 0:17for? Writing polite emails. Exactly. Writing
- 0:19polite emails to decline a calendar invite beat
- 0:23the gap between this raw planetary scale capacity
- 0:27and our everyday human utility. It is staggering.
- 0:30It is just a profound disconnect. It really is
- 0:33the modern paradox. I mean, people see the bleeding
- 0:35edge benchmarks. They read the headlines and
- 0:37they just shrug because it doesn't translate
- 0:39to their Tuesday morning workload, you know.
- 0:42Welcome to the deep dive. Today we are looking
- 0:45at Google's Gemini 3 .1 Pro. It was officially
- 0:48released in February 2026. That's right. And
- 0:51just for context, it scored a 77 .1 % on ARC
- 0:55-AGI2. Which is huge. Yeah. That is an advanced
- 0:58testing framework designed to measure an AI's
- 1:01true reasoning skills rather than just its ability
- 1:04to memorize data. But listen, we are entirely
- 1:06skipping the complex developer setups today.
- 1:08That is our real mission here. We have combed
- 1:11through the sources to find seven copy paste
- 1:14workflows. These are practical systems that make
- 1:17the free tier of Gemini do your actual heavy
- 1:21lifting work. No coding required. Right, no coding.
- 1:23Just pure leverage. We are going to climb a ladder
- 1:26of complexity today. We will start at the bottom
- 1:28by organizing flat static data. Okay. Then we
- 1:32will structure the physical world, time, and
- 1:34geography. From there we move into building interactive
- 1:37digital tools. Sounds good. And finally at the
- 1:39top of the ladder we will analyze the messy,
- 1:42unpredictable reality of human communication.
- 1:44I love it. Let's jump right into step one. The
- 1:48ultimate corporate headache. Oh boy. Turning
- 1:50flat data into visual stories. Oh, absolutely.
- 1:54Monday morning rolls around. You have a massive
- 1:56CSV file full of sales data and you have to present
- 1:59it. The worst. CSV files are miserable to read.
- 2:03And manually turning them into a slide deck usually
- 2:05takes, what, four hours of nudging text boxes
- 2:07around? Exactly. So we start this workflow using
- 2:10Canvas mode. For anyone unfamiliar, Canvas mode
- 2:13is a split -screen workspace where the AI builds
- 2:16an editable user interface right next to your
- 2:18chat window. Right. You do not just get a wall
- 2:20of text. You prompt the model, upload your messy
- 2:24CSV files, and it generates a fully structured
- 2:27slide deck. right there on the screen. But I'm
- 2:30trying to picture this. Is it just spitting out
- 2:32like a bulleted text outline that I still have
- 2:35to manually copy and paste into PowerPoint? No,
- 2:38no. It builds the actual visual presentation.
- 2:41Wait, really? Yeah. It creates a bold cover slide
- 2:44highlighting the main finding. It builds a source
- 2:47slide for your data transparency. Wow. It generates
- 2:50body slides laying out the strong and weak points.
- 2:53It even crafts a surprise slide with an unexpected
- 2:55insight it found in the numbers followed by a
- 2:58cool closing direction. That alone saves hours
- 3:01of formatting friction. But if you're presenting
- 3:03this to clients, it needs to look like it belongs
- 3:06to your company. It even handles the branding.
- 3:08Yeah. You just dictate your company's color hex
- 3:10codes in the prompt, you set the specific font
- 3:13styles, and the system picks the right visual
- 3:16charts automatically. So it decides the charts.
- 3:19It knows to use bar charts for comparing regional
- 3:21data or line charts for showing trends over time.
- 3:25Does the model actually understand visual hierarchy
- 3:28or is it just formatting text based on a preset
- 3:31template? Think of it this way. Gemini is not
- 3:33just looking at the words in your file. It mathematically
- 3:36maps your data points to establish design rules.
- 3:39Okay. If it sees a column adding up to 100%,
- 3:43that inherently triggers a geometric mechanism
- 3:45to visualize parts of a whole. It actually measures
- 3:49contrast ratios for the hex codes you provide
- 3:52to ensure the text is readable. So it acts as
- 3:54a geometric layout engine for your data. Spot
- 3:58-on you just review the final draft. It is absolutely
- 4:00perfect for internal reviews or you know weekly
- 4:02investor updates Okay, so we have successfully
- 4:05structured flat data visually. Let us climb one
- 4:08rung higher on the ladder. Let's do it We are
- 4:11going to apply that exact same structuring power
- 4:14to the physical world time logistics and geography.
- 4:19Travel planning. Yes. If you are listening to
- 4:21this and thinking, well, I have tried AI travel
- 4:23planners and they are completely terrible, you
- 4:25are right. They really are. They usually just
- 4:27spit out a generic top 10 list of tourist traps.
- 4:31Because the prompts people use are way too simple.
- 4:33Like, plan a trip to Portugal is not a strategy.
- 4:36Right. The fix here is giving Gemini a highly
- 4:39specific role. You do not ask it for a list.
- 4:42You tell it to act as a seasoned food writer
- 4:44spending four days in Porto, focusing only on
- 4:47local markets and authentic dining. I get that
- 4:50a role changes the tone of the output, but what
- 4:52about the actual logistics? Yeah. I hate when
- 4:54these things tell me to go to a cafe, then a
- 4:57museum across town, and then back to a restaurant
- 4:59near the first cafe. It is infuriating. That
- 5:02is where the workflow gets incredibly practical.
- 5:05In your prompt, you strictly demand that it groups
- 5:07all stops by geographical district. Oh, that's
- 5:09smart. This entirely avoids crossing the city
- 5:12back and forth. You force it to write a one -sentence
- 5:15justification for each stop. Then here is the
- 5:18killer feature. OK. You ask it to convert the
- 5:21daily mapped routes into shareable Google Maps
- 5:25URLs. That is brilliant. There is no manual copying
- 5:29and pasting of foreign addresses into your phone
- 5:32while you are standing on a sidewalk. Exactly.
- 5:35You just click the single link when you arrive
- 5:37at the airport and your entire day is routed.
- 5:39And it works for any scenario. Like what? A weekend
- 5:42traveling with toddlers, scouting remote coffee
- 5:45shops for deep work. You just swap the role on
- 5:47the prompt. Why does assigning a subjective role
- 5:50like a food writer? dramatically change the objective
- 5:54geographical output. Because roles act as strict
- 5:56negative constraints. When you assign a persona,
- 5:59the AI automatically filters out anything outside
- 6:02that specific persona's interests. I see. It
- 6:04stops processing data about historical monuments
- 6:07and only allocates its processing power to culinary
- 6:10data. A persona is just a sophisticated filter
- 6:12for geographical data. Exactly. It narrows the
- 6:15universe of options instantly. It puts blinders
- 6:18on the AI so it stays focused. Okay, let's keep
- 6:21climbing. If Gemini can map physical routes and
- 6:23filter the physical world, the next logical step
- 6:26is mapping digital workflows. We are going to
- 6:29stay inside Canvas mode for this one. Generating
- 6:32functional app prototypes. App prototypes. Yeah.
- 6:35You give Gemini a short, structured brief. In
- 6:38about three minutes, you get a working digital
- 6:41dashboard. Give me a concrete scenario. What
- 6:43kind of dashboard are we talking about? Imagine
- 6:45you manage a local co -working space. You need
- 6:48a dashboard that tracks daily desk bookings.
- 6:51It needs to handle guest check -ins. It has to
- 6:54show available meeting rooms in real time. You
- 6:56give it that brief, and it builds the interface,
- 6:59even populating it with realistic sample data,
- 7:01like names and times. You know, I have to admit,
- 7:04I still wrestle with prompt drift myself when
- 7:06building complex things. It happens to everyone.
- 7:09I ask for a dashboard, I try to fix one small
- 7:11thing, and by the third tweak, the AI forgets
- 7:14the original design entirely and the whole thing
- 7:15breaks. exact problem Canvas mode solves. Iteration
- 7:20is your safety net here. You do not need a perfect
- 7:23first prompt anymore. If a button is the wrong
- 7:26color, you highlight just that one specific button
- 7:28and tell the AI to change it. You do not regenerate
- 7:31or restart the whole app. You just refine the
- 7:33edges without breaking the core structure. Yes.
- 7:36You can specify exact visual cues block by block,
- 7:39make open desks emerald green, make booked desks
- 7:43amber. Nice. It creates a highly modular layout
- 7:46that works on both desktop and mobile views automatically.
- 7:49But can you actually test the logic flow or is
- 7:52it just a static mock -up, like a painted picture
- 7:54of a dashboard? You can actually simulate many
- 7:57scenarios. If you click book a desk in the preview,
- 8:00the UI will actively respond and change the state
- 8:02of that desk to booked. It is stateful. It builds
- 8:05a reactive environment, not just a painted picture.
- 8:07It is a massive shortcut for product teams. Developers
- 8:10can use these interactive prototypes as an immediate
- 8:13starting point instead of sketching on whiteboards.
- 8:16We just built a tool for internal use. Now let
- 8:19us push that exact capability outward. We are
- 8:22moving up to customer -facing tools. For this,
- 8:25we are moving out of Canvas and into Google AI
- 8:28Studio. Oh. For the listeners, AI Studio is Google's
- 8:32free developer playground for building and testing
- 8:34AI tools. We are going to create lead magnet
- 8:37widgets. By widgets, we mean standalone interactive
- 8:40tools you can actually embed on your own website
- 8:42to capture client interest. Exactly. Think of
- 8:44a B2B business creating an e -commerce ROI calculator.
- 8:48Right. You prompt AI Studio to build a tool with
- 8:50multiple active input fields, monthly ad spend,
- 8:54expected revenue, current conversion rate. Wait,
- 8:56with actual live inputs that the user can drag?
- 8:58real -time updating sliders the math recalculates
- 9:01instantly on the screen it even builds an email
- 9:04capture field at the bottom to lock in the lead
- 9:06right and whoa imagine deploying live interactive
- 9:10widgets in minutes without a front -end dev it
- 9:13is actually wild it builds immense trust with
- 9:15a potential user Interacting with a live calculator
- 9:18provides way more value than downloading some
- 9:20static PDF guide. And you can export the final
- 9:23code directly to GitHub. You can host it live
- 9:26on your site. This entire workflow completely
- 9:29bypasses traditional front -end development bottlenecks.
- 9:32But wait, how does it handle the underlying math
- 9:34without hallucinating the numbers? Good question.
- 9:38Language models are notoriously bad at reliable
- 9:40math because they just predict the next likely
- 9:43word. Because in this specific environment, it
- 9:46isn't predicting text for the answer. It actually
- 9:49writes and executes deterministic code based
- 9:52on the mathematical formulas you request in the
- 9:54prompt. It writes deterministic code to anchor
- 9:57the underlying logic. Exactly. So the math is
- 9:59flawless. Midrall sponsor read goes here. All
- 10:02right. We are back. We have mastered text, we
- 10:05have manipulated code, and we have built visual
- 10:07interfaces. We have. Now let us see how Gemini
- 10:10handles the absolute messy reality of human audio.
- 10:13We are talking about analyzing sales calls and
- 10:16team meetings. You stay right inside AI Studio
- 10:18for this. You upload a raw audio recording file
- 10:21directly, a messy client check -in, a chaotic
- 10:24weekly team sync. Audio is notoriously difficult
- 10:28to structure. Text is clean, but human speech
- 10:31is a disaster of interruptions and half -finished
- 10:34thoughts. Gemini handles the chaos natively.
- 10:37It automatically separates each speaker's lines.
- 10:39It labels exactly who said what, even if they
- 10:42interrupt each other. But if I am a sales director,
- 10:44I don't just want a raw transcript. No, of course
- 10:47not. Reading a 20 -page transcript of a meeting
- 10:49is practically useless. The output is far more
- 10:52advanced than transcription. It actually tracks
- 10:55emotional sentiment across the entire duration
- 10:58of the call. Then it generates a synthesized
- 11:00post -call review card. Almost like a senior
- 11:02manager sitting in the room giving you feedback.
- 11:05Exactly like that. It highlights your specific
- 11:07wins. Maybe it notes that you handled a hostile
- 11:10price objection perfectly at the 20 -minute mark,
- 11:12but it also flags your misses. It might point
- 11:15out that you jumped to pitching the pricing tier
- 11:17way too early in the conversation, and it pulls
- 11:20concrete audio timestamps and quotes to prove
- 11:23its point. Specialized software platforms that
- 11:25do this usually cost enterprise teams hundreds
- 11:28of dollars a month per user. And you can build
- 11:31a custom version for free in about 10 minutes.
- 11:34You can then share that exact grading workflow
- 11:37with your entire sales team, ensuring you have
- 11:39consistent, objective criteria for everyone.
- 11:43How nuanced is that sentiment tracking when multiple
- 11:46people are talking over each other? Human meetings
- 11:49get loud. It does not just read the transcribed
- 11:51words. It identifies individual vocal patterns.
- 11:55Like what? The pitch, the speed, the volume.
- 11:58It uses those to accurately map the emotional
- 12:01shifts of each specific speaker over time. It
- 12:04isolates emotional arcs for every individual
- 12:07in the room. It is brilliant for tracking your
- 12:09own communication patterns over time. You start
- 12:12to notice your own blind spots. Okay, so raw
- 12:14audio analysis is incredibly powerful for internal
- 12:17company. meetings. But what about analyzing public,
- 12:20highly polished video content? This workflow
- 12:23turns YouTube videos into highly polished written
- 12:25articles. And the best part is you do not need
- 12:27to download massive video files. You do not need
- 12:30third party transcription tool. You literally
- 12:32just use the public URL. Direct ingestion. You
- 12:35paste any public YouTube link directly into the
- 12:39prompt. Gemini automatically pulls the entire
- 12:41video content, the creator's description, and
- 12:44even the thumbnail image. But if I am honest,
- 12:46Every time I see a blog post that was clearly
- 12:49just a regurgitated YouTube transcript, it is
- 12:51a terrible read. Oh, they're usually awful. People
- 12:54speak very differently than they write. It never
- 12:56flows. That is usually the prompt's fault, not
- 12:58the AI's. The crucial step here is rigorously
- 13:02enforcing a style guide. Enforcing a style guide.
- 13:04You must dictate the exact tone. You dictate
- 13:08the sentence rhythm. You provide a list of cliche
- 13:11AI phrases to avoid. You must define the target
- 13:15audience You essentially make it act like a senior
- 13:18editorial writer. Yes. Not just transcription
- 13:20cleaner. You explicitly tell it to start the
- 13:23article with the single clearest takeaway. You
- 13:25tell it to move logically through the arguments.
- 13:27You command it to teach the reader, not just
- 13:29summarize what the guy in the video said. You
- 13:31feed it a URL. Does it literally watch the visual
- 13:33frames, or is it just scraping the hidden closed
- 13:36captions on the back end? This is the massive
- 13:38leap. It natively processes the visual video
- 13:41stream and the audio stream simultaneously without
- 13:44ever needing an intermediary text transcript.
- 13:47It sees the whiteboard diagrams, and it hears
- 13:50the explanation. Direct multimodal ingestion,
- 13:53completely bypassing the text middleman. It saves
- 13:56creators hours of repurposing work. You can turn
- 13:59a deeply researched video essay into a standalone
- 14:02high quality newsletter instantly. Digesting
- 14:05a single video perfectly is a great magic trick.
- 14:09but scaling that depth of analysis to an entire
- 14:12content ecosystem. That is the ultimate test
- 14:15of this system. Auditing entire YouTube channels.
- 14:17This final workflow is incredibly potent for
- 14:19strategists. You simply provide a YouTube channel
- 14:22handle. And Jim and I build a comprehensive diagnostic
- 14:25card for the whole brand. It checks the most
- 14:27recent upload batches. It pulls current viewership
- 14:29data. Right. It issues formal grades on the channel's
- 14:32market positioning, its posting cadence, and
- 14:34its audience response rate. It literally charts
- 14:37the growth curve. The sources mentioned a specific
- 14:39channel called AI Fire as a case study for this.
- 14:42Yes, the AI Fire example is perfect. Gemini analyzed
- 14:46the channel and gave it a C plus for positioning.
- 14:48Ouch. The feedback was brutal, but it was fair.
- 14:52It noted that the content was way too broad,
- 14:55which led to fragmented viewership. But it gave
- 14:58the channel an A - for posting cadence, praising
- 15:01their strong publishing systems. High -end media
- 15:03consultants easily charge thousands of dollars
- 15:06for that exact kind of strategic audit. Gemini
- 15:08executes it in about four minutes. It identifies
- 15:11precisely which video formats pull loyal viewers
- 15:14in and which formats drag the channel's overall
- 15:17performance down. Right. It uses actual historical
- 15:20video titles to give you concrete feedback. And
- 15:22it prescribes the next steps to fix the grades.
- 15:25It outputs three immediate quick wins, it suggests
- 15:28three long -term structural changes, and it pitches
- 15:31five highly specific new video ideas based entirely
- 15:35on empirical data. But let me challenge that.
- 15:37Are these letter grades purely arbitrary or are
- 15:39they anchored in something real? Fair question.
- 15:41I know AI loves to just invent authoritative
- 15:44sounding grades to please the user. They are
- 15:46not arbitrary at all. They are directly calculated
- 15:48from the channel's actual empirical data cross
- 15:51-referenced with historical audience retention
- 15:53patterns across the platform. The grades are
- 15:55strictly anchored in historical performance metrics.
- 15:58It is an objective reality check, not just an
- 16:00AI making educated guesses. We have looked at
- 16:03seven incredibly disparate workflows today. We
- 16:06went from geometric slide decks to travel logistics
- 16:10to code generation to channel audits. We covered
- 16:13a lot of ground. Synthesizing underlying engine
- 16:15that makes all of this function so well is critical.
- 16:18It really all comes down to mastering one big
- 16:21idea. Two -sec silence. The four -part prompt
- 16:24structure. Let's unpack this framework slowly
- 16:26because this is the engine. It is the absolute
- 16:28difference between generating generic garbage
- 16:31and building highly usable assets. Part one.
- 16:34Roll. You must tell the AI its job. Are you a
- 16:37minimalist product designer? Are you a seasoned
- 16:39editorial writer? Are you a B2B sales director?
- 16:42The role sets the entire intellectual approach
- 16:44for the task. Part two. Input. Give it the specific
- 16:48messy data, a CSV file, a YouTube URL, a chaotic
- 16:52audio recording. Without specific input, the
- 16:55system is just hallucinating in the dark. Part
- 16:57three. Output format. Tell it exactly what to
- 17:01build. Do not just say, make a thing out of this
- 17:04data. Right. Ask for a 10 slide deck. Ask for
- 17:07an interactive widget with live sliders. Ask
- 17:10for a diagnostic report card. You have to force
- 17:12its reasoning into a highly specific container.
- 17:15Part four, rules. Establish your negative constraints.
- 17:21Use these exact brand colors. Write at an eighth
- 17:23grade reading level. Never use the word synergy.
- 17:27Rules are what make the final output actually
- 17:29feel like it belongs to you. The sources used
- 17:32an interesting phrase. They said building prompts
- 17:34this way is like stacking Lego blocks of data.
- 17:37Yeah, I like that. You assemble these four distinct
- 17:39pieces, you click them together, and you build
- 17:40whatever machine you need for the day. I see
- 17:42what they mean, but I actually think it is a
- 17:44bit more dynamic than just stacking blocks. Yeah,
- 17:46honestly, I like to think of it more like setting
- 17:48up bowling bumpers for the AI. Bowling bumpers.
- 17:51Yeah. The role and the rules are the bumpers.
- 17:53Yeah. You are forcing the AI's massive processing
- 17:56power straight down the lane to the exact output
- 17:58format you want. Oh, that makes sense. If you
- 18:00set the bumpers correctly, the AI physically
- 18:03cannot roll off into the gutter of generic hallucinations.
- 18:06That is a much better way to look at it. Without
- 18:08those bumpers, that ball goes absolutely everywhere.
- 18:11Exactly. My advice to anyone listening is to
- 18:14pick just one of these workflows today. Try the
- 18:18presentation builder or the travel itinerary.
- 18:21Just pick one. Run your prompt. look critically
- 18:24at what it gives you, adjust your bumper rules,
- 18:27and run it again. In three quick rounds of iteration,
- 18:30you will have a reusable asset that saves you
- 18:32hours every single week. We leave you with this
- 18:36measured thought to mull over. We now live in
- 18:39a reality where a free AI tool can flawlessly
- 18:42mimic a senior graphic designer. It can mimic
- 18:45a seasoned travel writer. It can replicate a
- 18:47high -paid YouTube strategist, all conjured out
- 18:49of thin air from a simple four -part prompt,
- 18:51beat. if the machine can generate the perfect
- 18:54answer on demand at zero marginal cost. What
- 18:57becomes the uniquely human skill in the workplace
- 19:00of tomorrow? Beat. Perhaps the value shifts entirely
- 19:04away from answering things correctly toward asking
- 19:07the right questions.