Why AI doesn't recommend your startup (and the first 90 days to fix it)

The short answer

If you launched recently and AI assistants never mention you when buyers ask for options in your category, you are not unusual. You are the default. In our own checks of 50 software brands, every one of the 30 brands that had launched in the previous month scored 0: not named once in any answer to their own category's buyer questions.

That is not because the product is bad or the website is broken. AI answers to buying questions are built from what several trusted sources already say about a category. A new product has, by definition, almost no presence in those sources yet. The assistant cannot recommend what it cannot find described anywhere it trusts.

The good news is that the reasons are specific and fixable, and most of the fixes cost time rather than money. This guide explains what is going on and lays out a realistic first 90 days. It is written for founders and small teams who want their product to show up when someone asks an assistant for "the best [category] for [audience]" or "[market leader] alternatives".

What our own data shows

We ran unbranded buyer questions ("best [category]", "[competitor] alternatives", "X vs Y") through AI search for 50 software brands. These are the only numbers we quote anywhere on this site, and we quote them as they are.

• 39 of 50 brands (78%) were never named in any answer to their own category's buyer questions.
• All 30 brands that had launched in the previous month scored 0.
• Being known is not enough: an established e-signature tool scored 52/100 and an AI voice tool scored 37/100.
• In 33 of 50 cases the most cited source was a competitor's own website: its "vs" pages, docs and help centre.
• Review sites were the most cited source for only 6 of 50 brands.

Reason 1: the answer is built from sources you are not in yet

For buying questions, many assistants run a web search behind the scenes, read a small set of pages and then summarise what those pages agree on. The final answer usually names three to five options. A product that appears across several of the retrieved pages is far more likely to make that list than one that appears on a single page, even a very good one.

For a new startup the retrieved pages are typically roundups written before you existed, competitors' comparison and alternatives pages, documentation, review grids and community threads. You are absent from all of them. In our data, the most cited source in 33 of 50 cases was a competitor's own website, so in practice the incumbents' "vs" pages are often writing the description of your market.

Reason 2: your category is not stated plainly enough

Before an assistant ranks anything, it has to decide what kind of product the buyer wants and which products belong to that category. Early-stage sites often lead with a vision line ("the operating system for modern teams") instead of a category. If the model cannot tell whether you are a CRM, a project tool or an analytics product, the easiest outcome is to leave you out.

This is the cheapest problem to fix and one of the most common. Write one sentence of the form "[Product] is a [category] for [audience]", using a category name buyers already search for, and use it everywhere: title tag, H1, meta description, about page, directory listings and social bios.

Reason 3: the facts buyers ask about are hard to quote

• No public pricing. When a buyer asks for the cheapest option or a tool under a budget, products without a stated price are easy to skip.
• Docs or help articles behind a login. A gated page cannot be read or cited.
• Key copy rendered only by client-side scripts, so a crawler fetching the page on demand sees an empty shell.
• No FAQ answering the questions buyers actually ask: who it is for, what it costs, what it integrates with, what its limits are.

Reason 4: fame is not the whole story, but you have none yet

Being known helps, but it is not a guarantee. In our set an established e-signature tool scored 52 out of 100 and an AI voice tool scored 37, so even products with real market presence were missing from many answers. For a startup with no coverage at all, the gap is wider: there is no training-era memory of the product and no third-party page to retrieve.

This is why the fix for a new brand looks different from the fix for an established one. An incumbent tunes position and accuracy. A new brand has to earn its first mention, and that almost always happens through sources it does not own.

What not to do in the first month

• Do not judge visibility by asking an assistant about your brand by name. That tests recall by someone who already knows you, not discovery by a buyer.
• Do not publish dozens of generic blog posts. Thin content is rarely cited, and the pages that answer buying questions are comparisons, alternatives lists, pricing and docs.
• Do not call yourself "the #1 [category]". Self-declared rankings carry little weight next to independent sources.
• Do not assume review sites are the whole game. In our data they were the most cited source for only 6 of 50 brands. Check which sources your category's answers actually use.
• Do not pay anyone who promises placement in organic AI answers. There is no paid slot to buy.

Days 1 to 14: measure and fix the basics

1. 01Write 10 to 20 unbranded buyer prompts for your category: best-of, alternatives to the market leader, two head-to-head comparisons and a few constraint prompts (price, team size, a specific integration or region).
2. 02Run each prompt more than once and record every brand named, in order, plus every source cited. Expect to see zero mentions of yourself; the point is the source list.
3. 03Count how often each cited domain appears. The top five or six are your work plan for the quarter.
4. 04Lock your category sentence and update your title tag, H1, meta description, about page and every profile you control.
5. 05Publish public pricing with numbers, make docs readable without a login, and check the homepage's first paragraph is visible with scripts disabled.
6. 06Add Organization and SoftwareApplication structured data, FAQPage markup on genuine FAQs, a clean sitemap and an llms.txt file. Our guide on llms.txt and structured data covers the details.

Days 15 to 45: publish the pages answers draw on

1. 01Write an honest "[Market leader] alternatives" page listing five to eight real options, yours included, with who each suits and current prices. For a small brand this is often the easiest prompt type to appear in.
2. 02Write one "[You] vs [most-named competitor]" page with a facts-only comparison table and a "choose them if, choose us if" section. Pick the competitor named most often in your baseline, not the one you think of as your rival.
3. 03Cover one constraint where you genuinely win, such as the cheapest option for a five-person team or the only tool with a specific integration. Niche prompts are where new products can be named first.
4. 04Date each page and commit to re-checking prices quarterly.

Days 30 to 75: earn presence on sources you do not own

This is the part that usually decides whether a new brand gets named. Work only from your source-frequency list, not from a generic list of places to be seen.

• Email the authors of the two or three most cited roundups with a short, honest note on who the product is best for and a demo account. Ask for accuracy and consideration, not a ranking.
• List the product on the directories and launch sites that actually appear in your answers, using the same category sentence everywhere.
• Answer real questions in the community threads that get cited, say plainly that you work on the product, and only mention it where it genuinely fits the question.
• Ask your first active customers for reviews on the review platforms that appear in your category's answers, not on every platform.
• If a competitor's comparison page misstates your pricing or features, publish your own accurate comparison so answers have a second source.

Days 60 to 90: re-measure and narrow your focus

1. 01Re-run the exact prompt set from day one. Do not change prompts mid-quarter or the comparison is meaningless.
2. 02Look first at which cited sources now mention you. That usually moves before mention rate does.
3. 03If you are now named for any prompt, read how you are described. Fix the category sentence anywhere the description drifts.
4. 04Double down on the prompt type where you appeared, usually alternatives or a constraint prompt, before going after broad best-of prompts dominated by incumbents.
5. 05Set a monthly cadence for the same prompt set from here on.

What realistic progress looks like

Expect weeks, not days. Answers that use live web search can change as soon as cited pages change and are re-crawled, but earning a place in those pages takes outreach and time. Anything that depends on a model's training data moves on a much slower cycle.

A realistic first milestone for a new startup is not "named in every answer". It is being named in one or two constraint or alternatives prompts, appearing on two or three of the most cited third-party pages, and being described accurately when you are named. From there, share of voice can grow prompt by prompt. Our guide on measuring AI visibility explains how to track mention rate, position and share of voice without fooling yourself.

A worked example

Imagine a two-person team that has just launched an invoicing tool for freelance designers. Their baseline of 15 prompts names them zero times. The most cited sources are the market leader's own alternatives page, one roundup of invoicing apps from last year and a freelancer forum thread.

Their 90-day plan is narrow on purpose: change the homepage line from a tagline to "invoicing software for freelance designers", publish prices, write an honest alternatives page and one comparison against the most-named tool, pitch the roundup author with the designer angle, and answer two questions in the forum thread while disclosing who they are. They ignore broad "best invoicing software" prompts for now and track the constraint prompt "invoicing app for freelance designers", where a specific, well-described product has the best chance of being named first.

Where ShownInAI fits

You can do all of this by hand with a spreadsheet. If you want the baseline done for you, the $49 report runs 10 unbranded buyer prompts for your domain and returns a 0 to 100 score, the competitors named instead of you, the sources cited and the first fix to make. The $299 GEO audit covers 20 queries, a competitor comparison and a 30-day fix plan, and tracking plans from $29 a month re-run the same prompt set on a schedule so you can see the day-90 comparison without redoing the work.

For the full list of checks behind this plan, see our GEO checklist. For how to write the comparison and alternatives pages in the second phase, see comparison pages for AI search.