GEO for Small Business: When It Pays and What to DIY

GEO pays for most small online brands that sell something people research before buying, if you take it in stages. Spend an afternoon checking that your platform is not blocking AI crawlers and logging what engines get wrong about you. Then measure and fix it yourself for a quarter. Pay someone only if your fixes stop changing what the engines say. (Run a local service business? Start with your Google Business Profile and reviews.)

Where a small brand can win AI answers

AI engines decide which brand to name question by question, so a store that answers one narrow, constraint-heavy question better than anyone else can be named next to far bigger competitors. That gap is open now, and it rewards brands that can say exactly who they sell to and where they ship.

Our read on why ChatGPT names competitors puts it this way: "This is one of the few moments in the last decade where being small and specific is an advantage rather than a handicap. It won't last." The same piece notes that "Amazon is weak on those long-tail, multi-constraint questions."

Take a small Dutch brand selling refillable natural deodorant in the Netherlands and Germany. It will not be named for "best deodorant". It has a real chance at "What is the best aluminium-free deodorant for heavy sweating that does not mark dark shirts and ships to Germany? List brands and cite your sources." Every constraint in that question removes a big competitor.

How big is the prize today? Smaller than organic search, and closer to the sale. Shopify's Q1 2026 data says "More than half of AI-referred sessions start on product pages, compared to 20% for organic search," and the same report says "organic search remains the dominant discovery channel." To see your own number, filter your Shopify or GA4 referrers for chatgpt.com, perplexity.ai and gemini.google.com. Treat it as a floor: our GA4 setup guide explains why, and on stores we work with, a checkout survey asking how customers found you consistently reports AI discovery at a multiple of what the channel data shows.

GEO is a weaker bet in two cases. Lilach Bullock's small-business guide is blunt about the first: "If your business depends on impulse purchases, GEO is less relevant." The second is a category where even your narrowest questions return only marketplaces.

The afternoon check: is your platform deciding for you?

Before any content work, find out whether the crawlers that feed AI search can read your store. On a small site that is often decided by a platform checkbox, a CDN default or a robots.txt rule pasted in years ago. Here is what each common setting actually switches off.

Two kinds of crawler matter, and they are separate. Training crawlers collect text to build future models. Search crawlers fetch pages so an engine can answer and cite today. OpenAI documents them as different bots, and its crawler overview says sites that block OAI-SearchBot "will not be shown in ChatGPT search answers, though can still appear as navigational links." The most common technical error we find in audits is exactly this mix-up: a site allows GPTBot and assumes that makes it eligible for ChatGPT Search.

Squarespace

Squarespace's AI setting is a checkbox labelled "Block known artificial intelligence crawlers" in the Crawlers panel. It is off by default, and Squarespace explains that it leaves it off so as not to hurt traffic "by excluding it from chat answers and sources." Ticking it writes a list of named bots into your robots.txt.

The list includes GPTBot, ClaudeBot, Google-Extended and DuckAssistBot. It does not include OAI-SearchBot, ChatGPT-User, PerplexityBot or Claude-SearchBot. So a founder who ticked the box to keep product copy out of model training has not opted out of ChatGPT search or Perplexity by doing so. The box does cost something in Gemini: Google says Google-Extended controls training and also "grounding (providing content from the Google Search index to the model at prompt time to improve factuality and relevancy) in Gemini Apps". Google adds that it does not affect inclusion in Google Search.

Cloudflare

This applies if your own domain is proxied through your own Cloudflare account, which is uncommon on a standard Shopify storefront: Shopify's help page says it "doesn't recommend a proxy configuration in front of Shopify". Cloudflare's documentation, updated 1 July 2026, announced that from 15 September 2026 new domains get a default where training and agent bots are blocked on pages that display ads "and Search will remain allowed." A store without ad slots should be largely untouched by it, though Cloudflare's automated ad detection decides which pages count.

The rule to watch is older settings. The same page says "Mixed-purpose crawlers that combine Search and Training will also be blocked by all configurations to block AI training, including the legacy "Block AI bots" option." If you or a developer ever switched on a training block, check which crawlers your dashboard classes as mixed-purpose, because the documentation page does not name them.

Shopify

Shopify generates a default robots.txt, and changing it means editing the robots.txt.liquid template in your theme code, and Shopify's help page warns that "Shopify Support can't help with edits to the robots.txt.liquid file." Keep a copy of the original before you touch it. The thing to look for is an old block-all-AI snippet pasted in during the 2023–24 training debate. Whether it costs you AI search depends on the bots it names. If it lists OAI-SearchBot, ChatGPT-User, PerplexityBot or Claude-SearchBot, or blocks every bot with "User-agent: *", it removes you from those engines' live answers. If it names only training bots such as GPTBot or CCBot, it mainly affects training, and a Google-Extended line also affects Gemini grounding.

The same help page says blocking AI crawlers "at the /robots.txt or network layer affects only open-web discoverability." It adds that this does not stop Shopify Catalog sending product data "to the agentic storefronts that you've activated." So a store that is eligible for those channels and has switched them on can appear as a product in a shopping assistant and still be missing as a source in its search answers. Eligibility differs by channel and country; our Agentic Storefronts guide has the details.

The checklist

Open yourstore.com/robots.txt and search for GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, Claude-SearchBot and Google-Extended. A "Disallow: /" under any of them blocks that bot from the whole site. Check the setting your platform controls: the Crawlers panel on Squarespace, AI bot policies on Cloudflare, robots.txt.liquid on Shopify. Decide each crawler on purpose; allowing search crawlers while blocking training ones is a legitimate choice. View the source of a product page and search it for the price and one sentence of the description. Several AI crawlers execute little or no JavaScript, so if the details are not in that raw HTML, those crawlers see an empty page. Ask ChatGPT, Perplexity, Gemini and Google AI Overviews what you sell, where you ship and what your best seller costs. Write down every wrong fact, with the engine and the date.

Keep that list of wrong facts: it is what you fix first, before any visibility score. In the field notes we published on one of our own brands, the brand appeared in every case and the problems were wrong facts, which is why "correction precedes amplification in every engagement we run."

How many hours a month does DIY GEO take?

Weekly measuring alone takes 5.8 to 8.7 hours a month at LLM Pulse's planning rate, and fixing comes on top. That is at or above the 3 to 5 hours HubSpot says is enough to keep GEO basics in-house, and it uses most of Lilach Bullock's two to three hours a week, which is 8.7 to 13 hours a month.

HubSpot's small-business guide, written mainly for local service businesses, lists "You have 3-5 hours per month to dedicate to optimization" as a condition for handling the basics in-house. Lilach Bullock's guide says "One person can do this if they have two to three hours per week." Neither shows its working, so we built the measuring part from the bottom.

Our guide to getting recommended by ChatGPT tells founders to write ten real buyer questions, run each through four engines, log every brand and source, and repeat weekly. That is 40 checks a week, or about 173 answers a month. LLM Pulse's audit guide says to plan for "two to three minutes per prompt-engine combination," which puts it at 5.8 hours a month at the fast end and 8.7 hours a month of measuring at the slow end. The same guide's worked example, "three to four hours" for 30 prompts on 5 engines, works out to 1.2 to 1.6 minutes a check, or 3.5 to 4.6 hours a month for this panel. Plan on the higher figure until you have timed your own first week. Our guide to re-measuring cadence counts four-week windows instead of calendar months, which is why its figures are slightly lower.

If you only have Bullock's hours, measure every second week rather than cutting questions. That halves the logging to 2.9 to 4.3 hours a month, still gives each question two runs in every four-week window, and leaves 4.3 to 10.1 hours a month for fixes.

Two caveats on doing it by hand. Answers in your own logged-in browser are shaped by your history, so use a private window and keep the question wording frozen. And for scale: a founder logging about 173 answers a month collects roughly as many as our entry retainer, which logs "4 engines · 1 run each · 152 logged captures per month" across 38 buyer queries rather than 10. Paying buys more questions under fixed capture conditions, plus someone acting on what the logs show.

What to fix during the DIY quarter

Fix in this order: wrong facts first, then pages that answer your narrow questions directly, then the outside sources engines quote. Expect the first two to show within weeks and the third to take months, so judge the quarter on corrected errors and on which questions now name you.

HubSpot says small businesses "typically start seeing GEO results within 4-8 weeks of implementation". That fits the on-site part of the work. Our timeline separates the layers: access fixes land in days, content starts being retrieved in weeks, and the third-party layer that drives recommendations "shows consistent movement at three to six months." Expecting recommendations in eight weeks is the fastest way to give up just before they arrive.

For the deodorant brand, a quarter could look like this. Month one: correct whatever the afternoon check found, such as a wrong list of shipping countries or an old price, then rewrite the product page to state what buyers ask about, such as aluminium-free and whether it marks clothing, in plain sentences near the top. Month two: publish one page answering the heavy-sweating question honestly, including when a different product is the better choice. Month three: ask recent customers for detailed reviews and get the brand into the comparison lists and forums your questions keep citing, without astroturfing.

When does paying for GEO make sense for a small brand?

Pay for help when one of three things is true after your own quarter: engines still name competitors on questions you should own, you have errors you cannot trace to a source, or the hours cost you more than the help would. Then check the price against your margin before you sign.

Competitors still being named is the hardest of the three to fix yourself. Once access, accuracy and your own pages are fixed, the lever left is what other sites say about you. As we put it in our comparison of GEO and SEO budgets, "The work that moves AI citations doesn't happen on your website." Getting into the forums and comparison lists engines cite means weeks of emails and follow-ups, which is hard to do in spare evenings.

The smallest paid step is a one-time check. Our price list runs "from a €950 one-time report to €5,500/month retainers": the €950 AI Answer Integrity Report if the problem is wrong facts, or the €1,500 AI Visibility Audit, which "is credited in full against GEO Foundations if you continue within 30 days," if you need a full baseline. The entry GEO retainer is €2,200 a month with a three-month minimum. All prices exclude VAT.

Here is the arithmetic for the deodorant brand. Assume a €30 refill bundle that costs €12 to produce and deliver, so each extra order contributes €18. The integrity report pays back after about 53 extra orders and the audit after about 83, once. The entry retainer needs about 122 extra orders every month. Set that against your own time: at an assumed €75 an hour, Bullock's 8.7 to 13 hours is €650 to €975 a month. Swap in your own margin and hourly rate. If your AI referrals and your checkout survey cannot plausibly reach that order count within two quarters, keep doing it yourself. Our breakdown of what GEO costs walks through the same method for larger budgets.

Do not pay yet if your buyers do not research before buying, if your narrowest questions return only marketplaces, or if you cannot give it a full quarter. A brand in any of those spots should do the afternoon check, fix what is wrong, and put the rest of the budget into creative.

Your first week

Today, open your robots.txt and your platform's crawler setting and decide each AI crawler on purpose. Next, ask four engines the basic facts about you and log every wrong one. Then write ten buyer questions, freeze the wording and run them once in a private window.

That first run is a data point, not a baseline. Six to eight weeks of weekly runs gives you a baseline you can defend.