You have decided to do this. Somewhere between the last board meeting and this morning, GEO stopped being a someday item and became a this-month item. The problem is nobody handed you a day one. You have a blank calendar, a vague sense that ChatGPT and Perplexity matter, and thirty days before someone asks what changed.
Here is what those thirty days actually cost you if you do nothing: not a catastrophe. Most brands in most categories are doing close to nothing on this front, which is exactly why thirty focused days puts you meaningfully ahead of the field rather than merely caught up to it. The real cost of waiting is not sudden, it compounds quietly. Every month you are absent from a comparison thread or a Reddit answer is a month that gap sits in an AI engine's retrieval index, or worse, its next training snapshot, slowly hardening an incomplete picture of your brand into something harder to correct later.
That is a solvable problem on a thirty-day clock, but one thing needs to be said before the plan starts, so you do not finish the month disappointed. Thirty days will move your visibility in the engines that retrieve live content: Perplexity, Tavily, Google AI Overviews, and AI Mode. It will not move your visibility inside ChatGPT's or Claude's underlying training data in any way you can measure this month. Those models update their core knowledge on a schedule measured in months, sometimes longer, and nothing you publish in the next thirty days changes what they already learned. If ChatGPT recommendations are your real target, this plan is still the correct first move, because everything you publish now becomes eligible for the next training pass. But that payoff arrives later and on a timeline you do not control. Say this out loud to whoever is expecting a ChatGPT number to jump by day thirty. It will not, and pretending otherwise is how these projects lose credibility around week five.
Before Week One: What You Need in Place
Two things need to exist before day one starts, or the whole month runs on guesswork.
The first is a prompt set: 15 to 20 real buyer-intent questions your prospects would plausibly type into an AI engine. Not keywords. Questions. "Best project management tool for a 20-person agency" reads differently to a retrieval engine than "project management software," and your baseline needs to reflect how people actually ask. If you don't have this list yet, build it from sales call transcripts, support tickets, and the actual review language customers use on G2 or Capterra. Twenty minutes with a search function beats an hour of guessing in a conference room.
The second is a way to check where you currently stand. That can be a paid tracking tool, or it can be you, manually, running your prompt list through ChatGPT, Perplexity, Gemini, and Google AI Mode and recording what comes back. Manual works fine for a first baseline. It does not scale past a handful of prompts checked repeatedly, but for thirty days and twenty prompts, it is sufficient and free.
The Mistake Almost Everyone Makes on Day One
The most common way this plan fails before it starts is building a prompt list out of keywords instead of questions. "Project management software" is a keyword. Nobody types that into ChatGPT. "What's the best project management tool for a 20-person agency that bills hourly" is what someone actually asks, and it produces a completely different answer, because the engine now has to reason about company size, billing model, and category fit rather than just matching a term.
Write your prompts the way your prospects actually talk, including the messy, specific parts: their industry, their team size, their budget range, the competitor they're already using. Pull the exact phrasing from sales call recordings and support tickets if you have them. A prompt list built from real language outperforms one built from a marketing meeting every time, because it's the list an AI engine is actually being asked to answer.
One more thing worth doing before day one: write down, honestly, what "success" looks like at day thirty. Not a number pulled from a vendor's sales deck. A specific, achievable claim, such as "we appear in the top three results for at least half our prompt set on Perplexity and AI Overviews, with no factual errors in how we're described." Setting this now, before you've seen any results, keeps you from moving the goalposts later in either direction.
Week 1: Baseline and Audit
The first week produces no visible change. It produces the numbers everything else gets measured against, and a clear list of what is actually broken.
Run your baseline. Work through your prompt list across every engine you care about. Record, for each prompt: whether you appear at all, where you rank relative to competitors, and what the engine actually says about you. Screenshot everything. You will want this exact wording again in week four.
Audit your structural presence. Check whether your Wikidata entry exists and is accurate. Check whether your G2, Capterra, and Trustpilot profiles are complete, correctly categorized, and carry your current logo and description. Check your site for basic Organization and Product schema markup. None of this alone moves rankings, but its absence is a real ceiling on everything else you do this month.
Hunt for entity confusion. Search your brand name alongside "vs," "review," and "alternative" across ChatGPT, Perplexity, and Google AI Overviews. Note anywhere an engine confuses you with a similarly named company, cites an outdated fact, or misstates your category. This is more common than founders expect, and it is worth catching in week one rather than discovering it in week four when there is no time left to fix it.
Pick three competitors to track alongside yourself. A single number in isolation tells you little. Knowing you moved from position four to position two while a competitor stayed flat tells you the movement was real, not noise.
Week 2: Fix the Structural Foundation
Week two is unglamorous and it matters more than any single piece of content you'll publish this month.
Complete your schema markup. If your audit in week one found gaps, close them now: Organization schema with your logo, founding date, and social profiles; Product schema with pricing and category; FAQ schema on any page that already answers common questions in prose. This is a few hours of engineering work, not a strategy.
Fix your review platform profiles. A G2 or Capterra profile with a generic description and a stale logo reads as neglected to both humans and the engines that crawl it. Rewrite the description in specific, category-clear language. Update every stale field.
Address anything Wikidata-eligible. If you have a real, notable presence and no Wikidata entry, creating one is a same-day task with an outsized effect on how confidently AI engines describe who you are. If you don't clear that bar yet, don't force it. A rejected or thin entry does more harm than no entry at all.
Publish one honest comparison page. "[Your Brand] vs [Top Competitor]," written to be genuinely useful to someone deciding between the two, not a thinly disguised sales pitch. AI engines retrieve from these pages constantly when users ask comparative questions, and a fair one earns more trust, and more citations, than a slanted one.
Week 3: Feed the Retrieval Engines
This is the week where Perplexity, Tavily, and AI Overviews start to notice you, because these systems retrieve fresh content on a rolling basis, not a fixed training schedule.
Publish two or three genuinely citable pieces. A specific data point from your own usage or a small survey. A clear definitional page for a term your category argues about. A how-to that answers one narrow question completely instead of ten questions shallowly. Specificity is what gets pulled into an AI-generated answer. Vague thought leadership rarely does.
Generate real reviews. Aim for five to ten new reviews on your primary platform this week, from actual customers describing actual use cases in their own words. Reviews that read like natural language, not marketing copy, are more useful to a retrieval system and more convincing to a human reading them.
Participate in two or three relevant community threads, on Reddit, Quora, or wherever your buyers actually discuss this category. Be helpful first. A single well-placed, honest answer in an active thread outperforms a dozen promotional posts that get flagged or ignored.
Start one earned-coverage outreach, even though it will not land inside this month. A journalist reply, a podcast pitch, an analyst briefing request. This guide is about the thirty days you can control. Earned media runs on a much longer clock (see Earned Media as a GEO Channel) but the outreach itself costs almost nothing to start now.
A quick note on what "citable" actually means in practice, since it's the part teams get wrong most often. A blog post titled "5 Tips for Better Project Management" is not citable in any useful sense. It's generic, and an engine has no reason to pull your version over the thousand similar posts already in its index. A page titled "How Long Does It Take to Onboard a 20-Person Team Onto [Category] Software" that answers with an actual range, a specific process, and a named example is citable, because it answers one question completely and specifically enough that quoting it is more useful than paraphrasing it. Aim for one page like that per week, not ten shallow ones per month.
Day 15: A Realistic Gut Check
Halfway through the month is a natural point to notice whether this is actually working, and it's worth pausing here rather than waiting for the week four report to find out.
If your week one audit turned up entity confusion or a genuinely broken review profile, day 15 is when you should see those specific items resolved, not just started. If they aren't, that's the thing to fix before adding anything new, since new content built on top of a confused entity or an abandoned review profile does less work than it should.
If nothing at all has changed by day 15, the likely cause isn't the plan, it's usually one of three things: the prompt list is still keyword-shaped rather than question-shaped, the content published so far is generic rather than specific, or the category is simply saturated with a small number of dominant, deeply established players. The first two are fixable inside the remaining two weeks. The third is worth knowing now rather than discovering at day thirty (more on that in What to Do When Nothing Is Working).
Week 4: Re-Measure and Set the Cadence
Re-run the exact baseline from week one, same prompts, same engines. Compare wording, not just position. Sometimes an engine still ranks you the same but now describes you correctly where it didn't before, which is real progress the position number alone won't show.
Report the split honestly. Expect measurable movement on Perplexity, Tavily, and AI Overviews. Expect flat numbers on ChatGPT and Claude, and say so plainly in whatever readout you give. A flat ChatGPT number after thirty days is not a failure of this plan. It is the plan working exactly as it should, given how those models actually update.
Build the ongoing rhythm from what worked. Which content type got picked up fastest. Which review platform moved the needle. Which community actually engaged. Thirty days is enough to learn your category's specific pattern. Keep doing more of whatever worked and drop what didn't, rather than repeating the same fixed checklist every month regardless of results.
What Thirty Days Will Not Fix
A few things are worth naming honestly before you start, so nobody is blindsided in week five.
Schema markup and structural fixes are necessary, not sufficient. They remove obstacles; they don't manufacture authority you haven't earned yet. If your category has three well-established incumbents with a decade of press coverage and reviews, thirty days closes some of the gap and does not erase it.
Entity confusion, if your audit turns any up, sometimes takes longer than a month to fully resolve, particularly if it involves getting a third-party source corrected or waiting for a Wikidata edit to propagate.
And real earned media, the kind that meaningfully shapes how AI engines describe your category, runs on a quarters-long timeline no thirty-day sprint can compress. Treat this plan as the foundation that makes the next ninety days more effective, not as a complete strategy on its own.