Which Pillar Of AI Engine Visibility To Fund First
Two questions about your category, how crowded it is and who buys, set whether your site, mentions or Reddit gets the money first.
On this page
Empact Newsletter
One email a month on what is working in AI search and SaaS go-to-market.
AI search visibility rests on three pillars, and each one is a line in your budget. Your own site is one, mentions of your product on other people’s pages are another, and the Reddit threads and communities where buyers compare notes are the third. Most advice tells you to fund all three. The rest gives you one fixed order, and it usually starts with the website.
With one budget, the order is the decision. Whatever you fund first is what you will be asked about a quarter later, when someone wants to know whether any of it is working. Two questions about your own category set that order: how crowded the category is, and who your buyer is. You can answer both in an afternoon.
Empact Partners runs all three lines for SaaS companies that want ChatGPT, Google AI Mode and Claude to name them in their answers. The roadmap that opens each partnership puts the lines in order, and these two questions are what we look at first.
How Crowded Is Your Category?
Open your category on G2 and note the listing count printed at the top. Then ask ChatGPT and Google AI Mode the questions your sales team hears most, and count the names that keep coming back. Under about 50 listings, treat the category as narrow. Past about 250, it is crowded.
Those bands come from our own run of three buying questions in 110 G2 software categories on 28 October 2026. The categories ran from niches of a few dozen products to fields of over a thousand. In every size band, the answers named a median of 14 vendors per category.1 What changes with size is everything around those 14:
Scroll horizontally to see all columns when the table is wider than this page.
| What we measured | Under 50 listings | 50 to 249 | 250 or more |
|---|---|---|---|
| Vendors named per 100 products listed | 39.3 | 11.8 | 2.5 |
| AI Mode answers citing a community page | 9.5% | 25.3% | 47.6% |
| AI Mode answers citing a publisher | 39.2% | 37.3% | 51.4% |
| AI Mode answers citing any vendor’s own site | 89.2% | 88.7% | 88.6% |
Fourteen names leave room in a field of a few dozen products and very little in a field of several hundred. The bigger change is in what gets cited. Community pages, Reddit and YouTube among them, showed up in about half of AI Mode’s answers in crowded categories and in fewer than one in ten small ones.
Some vendor’s site gets cited in almost every answer, and it is rarely yours. The Empact Panel, our own tracker, re-asks a partner’s buyer questions on ChatGPT, Google AI Mode and Claude every week and logs every cited page. Here is one software partner, in a category where it tracks 304 brands, over 30 days:

The card shows a blog with dozens of cited posts in a crowded field, still holding a tenth of a percent of what the answers cite. Across the 11 business software partners in our Panel cut, the partner’s own site took a median 3.2% of ChatGPT’s citations and 5.1% of Google AI Mode’s.2
In a crowded category, even a well-cited site is a few percent of what the answers read.
Vendor sites as a group are everywhere in the answers, and any one vendor’s part of them is small. So in a crowded category we would put the first money off your site, on the pages the answers already read. Judge it by whether those pages start naming you.
The site still has to be right. The answers that do cite it repeat what it says about what the product does, for whom, what it costs and how it compares. Correcting those pages is a prerequisite rather than a budget line. Growing the blog is what waits.
Telling a team in a crowded category to stop growing the blog is a hard sell, because the blog is the line they control. But when the answers already cite hundreds of other pages, one more post is a rounding error. The listicles and threads those answers lean on are where the first quarter’s money belongs.
Shanal Govender on LinkedInOpens in a new tabA narrow category turns that around. With about four vendors named for every ten products listed, a product the answers can describe has a real chance of being named. So the first money goes to use-case pages, one for each role, industry or job your product does, starting with the ones your rivals have not written.
Who Is Your Buyer?
The second question decides where the off-site money goes. Each kind of buyer checks a product in different places, and the answers to their questions cite those places. A placement that carries weight with one buyer can be invisible to another.
Reddit shows the spread most clearly. On 16 partner panels, we measured how many of Google AI Mode’s answers with sources cited at least one Reddit page. The share ran from 0.3% for a care provider that families choose to 32.1% for a developer tool, with business software in between:2
Google AI Mode Answers Citing Reddit, by Buyer
Answers with sources that cite at least one Reddit page, one bar per partner panel
View the data
| Label | Value |
|---|---|
| Developer tool 1 | 32.1% |
| Developer tool 2 | 16.3% |
| Business software 1 | 15.3% |
| Business software 2 | 14% |
| Business software 3 | 12.5% |
| Consumer app | 12.1% |
| Business software 4 | 11.4% |
| Business software 5 | 9.6% |
| Business software 6 | 9.3% |
| Business software 7 | 8.4% |
| Business software 8 | 4.6% |
| Business software 9 | 4% |
| Business software 10 | 2.2% |
| Business software 11 | 1.7% |
| Care provider 1 | 1.1% |
| Care provider 2 | 0.3% |
ChatGPT cited Reddit far less on the same panels. On no business software panel did Reddit reach more than 0.2% of ChatGPT’s citations, against a median 4.4% of AI Mode’s. So whether Reddit belongs in the first quarter is a question about your buyer and the engine they use, never a default.
The other pages split the same way. For software a business team buys, the answers lean on rival vendors’ own sites, big platforms’ help pages, tech publishers and review sites. For products a person picks, they lean on finance publishers for investors, and on associations, government pages and hospitals for families choosing care.
Profound’s study of 27 million citations shows the same pull across whole industries, with institutions at 30% of healthcare citations and 8% of software’s.3 So the mentions line in your budget is really a set of pages, and the set depends on who is asking:

To see your own set, check each source your buyers’ answers cite for a rival’s name and the absence of yours. Our AI Citation Gap Report finds the independent ones from your website, leaving rivals’ own sites out. It groups them by type of page and says the way into each. The report is free, and it is the first pass of an agent our partners get running in full:
Read Your Order Off the Table
Put your two answers together and the table gives you an order. Each row assumes your site already states the facts an answer needs. Treat each “first” as the line that gets the first and largest share for a quarter, as a test, with the second line behind it:
Scroll horizontally to see all columns when the table is wider than this page.
| Your category and buyer | Fund first | Fund second | Hold back on |
|---|---|---|---|
| Narrow, business buyers | Use-case pages on your own site | The review profiles and few listicles your answers cite | Reddit, until threads show up in your answers |
| Narrow, consumers or developers | Use-case pages on your own site | The publishers and threads your answers cite | A volume blog program |
| Crowded, business buyers | Placements on the listicles, review sites and trade pages your answers cite | Reddit threads, where your answers cite them | New blog posts |
| Crowded, consumers or developers | The publisher pages and Reddit threads your answers cite, Reddit only where they cite it | Listicles and review sites | New blog posts |
The table cannot settle everything. A category between 50 and 250 listings goes by whether the same few names fill every answer. A product sold to two kinds of buyer gets two sets of targets.
And none of these figures says what a placement will return, only where the answers look. That is why the first line is judged on your own readings before the second gets money.
What We Would Fund in the First Quarter
The row you land on is a hypothesis about your category. The first quarter tests it, and these steps get you from your two answers to the first money spent:
Write down your row.
Your listing band, your buyer, and the line the table funds first.
Collect your buyers’ answers.
Ask ChatGPT and Google AI Mode the questions your sales team hears most, and save every page each answer cites.
Sort the cited pages by owner.
Your site, rivals’ sites, review sites and listicles, publishers, Reddit and other communities, public bodies. Leave rivals’ own sites aside and mark the pages that name rivals and skip you. Those are your candidate placements.
Fix the cited pages that are yours.
Correct the facts, the price and the comparisons, and say plainly who the product is for.
Put most of the money on the first line.
Placements, threads or use-case pages, whichever your row says, rather than an even split across all three.
Ask the same questions every week.
Keep the wording fixed and compare which pages get cited, and whether you are named, against your first reading.
Our partnerships at Empact Partners open with an audit and a written roadmap, and the order of these three lines goes into it before any content is planned. We would read your buyers’ answers before setting it.
In a crowded category we would lead with Mention Building, add Reddit Marketing where your buyers’ answers cite threads, and keep the site work to the pages already cited. In a narrow one, the first quarter goes to use-case pages we write with your team. The work covers:
- Finding the listicles and review pages your buyers’ answers cite, and asking each editor to add your product, with what their readers would want to know about it
- Reddit threads and replies written to answer the person asking, inside each community’s rules, naming your product only where it belongs
- Your own pages rewritten so an engine can read them and quote them accurately, beginning with those already cited
- A weekly Empact Panel report on which of those placements, threads and pages the answers cite, with the next move for each
For the method on paper, our four-page playbook on growing AI visibility covers getting onto your buyers’ AI shortlists and telling what pipeline that brings:
Whichever row you are in, give it a quarter of weekly readings before you judge it. The first sign the order is working is the pages you funded appearing among the sources, and the next is your name appearing in the answers that cite them.
Frequently Asked Questions
How do I know whether my software category counts as crowded?
Look at two numbers. The listing count on your G2 category page is the first: under about 50 is narrow, past about 250 is crowded. The second is how often the same names return when you ask ChatGPT and Google AI Mode your sales team’s questions.
For a count between 50 and 250, we treat the category as crowded when the same few names fill every answer. When the names change from one question to the next, we treat it as narrow.
Do we pause blog posts while the money goes to mentions?
Slow them down rather than stopping. Keep the pages that answers already cite accurate, and keep writing what your sales team needs.
What waits is a program built on volume, because in a crowded category each new post joins hundreds of pages the answers can choose from. Put the saved money into placements on the pages your buyers’ answers already read. Come back to the blog once those placements are live and you can see what the answers cite next.
What if we sell to businesses and to consumers?
Run the second question twice, once per buyer, with the questions each one asks. Each buyer asks different questions, and the answers to them cite different pages.
You may end up with two sets of targets, review sites and trade pages for one, Reddit and publishers for the other. Fund the buyer that brings more revenue first, and add the other buyer’s pages once the first set shows up in the answers.
Does Reddit matter for B2B software in AI search?
Reddit’s weight depends on the engine and on your buyer. On the business software panels we read, ChatGPT cited Reddit in well under 1% of its citations, while Google AI Mode cited it more. Developer tools sit at the high end.
Before you fund threads, check whether the answers to your own buyers’ questions cite any. If they do, those threads are worth joining within each community’s rules. If they never do, put the money into the pages that are cited.
What would tell us we funded the wrong line first?
Two readings, a few weeks apart, on the same questions.
Say the pages you won places on start appearing among the cited sources while your product still is not named. Then you are on the right pages, and the next thing to check is how they describe you.
If the cited pages change and none of them are ones you worked on, the answers have moved elsewhere. Choose the targets again before more money goes in. Judge the order over a quarter, never a week.
Is our own site worth fixing if it is only a few percent of citations?
Yes, because the few answers that do cite it repeat what it says. A wrong price, a retired feature or a vague one-line description gets repeated to the buyer.
Fix the facts on the pages that are already cited, and make sure each page says what the product does, who it is for and how it compares. That work is small next to a content program, and it protects every placement you win elsewhere, since a buyer who clicks through lands there.
Do ChatGPT and Google AI Mode need different orders?
Often the order holds and the targets shift. In our category run, AI Mode’s answers in crowded categories cited community pages about half the time, while ChatGPT cited them in hardly any. ChatGPT cited far fewer pages per answer of any kind.
If most of your buyers use one engine, weight its cited pages more heavily when you choose the targets. Track the two engines separately so one does not hide the other.
Should a new product in a crowded category narrow its category instead?
Sometimes. If your product does one job better for a specific buyer, use-case pages for that job put you in a much smaller field. In fields that small, our run found about four vendors named for every ten products listed.
Test it with the questions that buyer asks first. If the answers still name the big category’s leaders, you are in the crowded case, and the off-site work comes first.
Should the budget split match the share of citations each kind of page gets?
No. Citation shares tell you where the answers look, not what a placement costs or how long it takes to win. A review profile is yours to correct, while a place on a well-cited listicle waits on its editor. Use the shares to rank the targets and to decide which line goes first, then fund that line properly before spreading money across all three.
What makes a use-case page worth funding first?
A use-case page worth funding is written for one role, industry, platform or job, in the words buyers use when they ask an AI engine for a recommendation. It says what the product does in that case and how it compares with the tool buyers usually start from.
The page also carries the facts an answer needs to repeat, such as price and the integrations that matter. A page that could describe any product in the category gives an answer nothing to pick you for.
Sources (3)
- Empact Partners, "AI answer run in 110 G2 software categories" (2026)Empact Partners' own data↩1three unbranded buying questions per category, ChatGPT with web search asked twice and Google AI Mode once, US English, 28 October 2026, 989 answers, banded by G2 listing count. Published in our SaaS category competition statistics. Re-read 16 November 2026.
- Empact Partners, "Empact Panel cut" (2026)Empact Partners' own data↩1↩2↩3every page cited in the latest two weekly collections after 9 October 2026 on 16 partner panels (11 business software, 2 developer tools, 1 consumer app, 2 care providers), ChatGPT and Google AI Mode, sorted by who owns the page. Read 16 November 2026.
- Profound, "Who Shapes AI Answers? Introducing Enhanced Citation Categories" (2026)↩18 January 2026. 27 million citations across ChatGPT, Gemini and Google AI Overviews. Read 16 November 2026.
Keep reading
GEO Should You Fund Your Website, Mentions or Reddit First?
Find the gap in your website, external mentions or user discussions, then choose the next job for your SaaS team.
GEO Does Ranking on Google’s First Page Get You Into AI Answers?
Rankings steady but clicks falling? Check which pages AI cites and decide whether to improve your pages or earn mentions elsewhere.
GEO How to Measure AI Visibility Engine by Engine
Separate brand mentions from source citations, compare engines fairly, and choose where your next quarter's AI-search work goes.

