Across 137 SaaS companies and 120 buying queries, the brands AI engines cite most have two things at once. They have high-volume presence on review directories, and they have strong topical authority on Google. Either signal alone gets you partway. Stacked together, they produce 5x more ChatGPT citations than authority alone and 2.4x more than directories alone. We call this the Compounding Rule.

Why we did this study

I get asked the same question by every SaaS founder I speak to right now. “How do I get cited by ChatGPT?” The honest answer is that nobody had measured it across enough companies to give a tactical answer. So we decided to.

EMGI’s original 150-company AI citation gap study measured ChatGPT citations and Google authority across the SaaS ecosystem. It surfaced the finding journalists picked up: 44% of SaaS brands with strong Google rankings have no ChatGPT visibility at all. The study told us the gap was real. It did not tell us what closes it. Half the picture was missing.

The missing half was directories. Reviews. Listings. The third-party authority surfaces that AI engines parse alongside the open web. So I asked Katya at Blastra if she would share crawl data from their listings management platform, because Blastra builds and tracks the most thorough SaaS directory presence I have come across. Twenty-three review and listing platforms. 139 SaaS brands. 2,095 individual listing checks. Review counts, ratings, category placements and comparison links captured for each one. The directory side of the AI corpus picture, fully mapped.

So we merged the two datasets. 137 SaaS companies overlap, which means we looked at the exact same companies on both sides. That overlap is what lets us answer the question I think every founder wants answered: how do directories and link authority combine to produce the citations AI engines hand out, and which lever moves more?

My lane is link building and topical authority. Katya’s lane is directory presence. We had a hunch the two compound. The data turned out to be more direct about it than we expected.

How we measured the Compounding Rule

Two datasets. 137 SaaS companies. Six categories. 120 buying queries. Three AI surfaces verified. Here is exactly how we built the picture, because credibility is the whole game when you publish original research.

The directory dataset (Blastra)

Blastra is a SaaS listings management platform that builds and maintains directory presence for B2B software companies. The platform’s crawl infrastructure surveyed 139 SaaS companies across 23 directory platforms and completed 2,095 individual listing checks. Every check captured the directory profile URL, the review count, the average review rating, the category placement Blastra found the brand listed under, the comparison and alternatives links the directory had set up, and an agent confidence score that flagged listings that looked thin or unclaimed.

The directories covered the entire spectrum of SaaS authority surfaces. Gartner Peer Insights for the enterprise editorial directories. G2, Capterra and TrustRadius for the high-volume review aggregators. SoftwareReviews and Crozdesk for the editorial-style platforms. PeerSpot, SoftwareAdvice and GetApp for the buyer-focused review listings. Trustpilot for general consumer trust signals. Crunchbase for company verification. AlternativeTo, SourceForge and F6S for community-driven listings. Plus nine vertical-specific directories that surfaced for individual categories. If a SaaS brand has any meaningful directory presence, this dataset has captured it.

For full transparency: the raw directory crawl data is available from Blastra for cross-validation, and we have offered the joined-data side of the analysis to anyone who wants to reproduce the audit. Three claims in this article (the CRM review-volume floor, the directory lift bars, and the five anonymised bottom-quadrant brands) involve joined-data calculations that sit between the two datasets. We are happy to share underlying numbers with directory operators, researchers, or SaaS founders who want to verify before quoting.

The AI citation dataset (EMGI)

We ran 120 buying-intent keywords spanning the same six SaaS categories. Each query went to ChatGPT search and to Google search with AI Overview detection in the same week. Per query, we recorded which brands ChatGPT named in its response, which brands appeared inside Google AI Overviews, and which brands ranked in organic SERPs from position 1 to 100. We also pulled brand-level totals: how many keywords each brand ranks for in organic Google, how much organic traffic value those rankings translate to, how many web mentions exist for each brand, and the sentiment skew of those mentions.

How we joined the two

137 Blastra companies matched 137 from our previous study. We looked at the exact same companies on both sides. That is the cleanest possible merge for this kind of research, and it is the reason we can talk about correlations with any confidence.

How we verified across surfaces

We did three more things to harden the findings. We ran live ChatGPT scrapes through our own scraping pipeline on seven buying queries spanning all six categories, captured in April. We ran the same prompts on Perplexity Sonar and Gemini 2.5 Flash to confirm the brand cluster appears across surfaces, not just on ChatGPT. And we ran per-cohort correlations to test whether the patterns hold inside each category, not just on the full-sample average. Correlation does not prove causation, and we are not claiming it does. We are showing the patterns that consistently appear, and the actions practitioners can take from them.

Do Google AI Overviews, ChatGPT, Perplexity and Gemini cite the same brands?

Google AI Overviews and ChatGPT do not just behave alike. They cite essentially the same brands. The two move in lockstep more than 9 times out of 10. We tested it. The same is true on Perplexity and Gemini.

Across our 137 brands, AIO appearances and ChatGPT citations correlate at 0.91. Practically, that means the brands surfaced in Google AI Overviews are the same brands ChatGPT cites more than 9 times in 10. One underlying authority signal is being read by two different surfaces.

The pattern holds inside every SaaS category we measured. CRM correlates at 0.97. Project Management at 0.96. Marketing Automation at 0.93. Dev Tools at 0.87. Analytics at 0.79. HR comes in lower at 0.59, which we read as smaller sample size and more closed buyer behaviour, since HR buyers rely heavily on RFPs and procurement rather than open search.

We then took the next step. We ran the same buying prompt against three AI engines simultaneously to see if the brand cluster matched.

When we ran “best CRM software for small business” through ChatGPT, the response named HubSpot, Zoho, Pipedrive, Freshsales, Insightly, Salesforce Essentials, Monday.com CRM, Nimble and ActiveCampaign. Every one of those brands also appears in Google AI Overviews for the same query.

We then ran the same prompt through Perplexity Sonar. It named HubSpot, Pipedrive, Zoho, Microsoft Dynamics, Freshsales and Bitrix24.

Then we tried Gemini 2.5 Flash. It named HubSpot, Zoho, Pipedrive, Freshsales, monday CRM, Less Annoying CRM, Salesforce Starter, Zendesk Sell, OnePageCRM and Bitrix24.

Three different AI engines. The same core brand cluster. They are reading the same authority signal layer.

The implication for SaaS founders is the most useful thing in this study. Stop optimising for ChatGPT separately from AIO. The work that lifts ChatGPT lifts all of them. The Compounding Rule is how AI search works.

What happens when you plot directory volume against topical authority?

Plot every SaaS brand on two axes, directory review volume and topical Google authority, and the Compounding Rule shows up immediately. The shape of the data is not subtle. It is a quadrant where one cell wins and the other three lose at predictable rates.

Using a median split on both axes (1,528 directory reviews and 10,699 Google rankings), every brand drops into one of four quadrants. The citation rate per quadrant is the entire article in one table.

The headline lift is in the top-right cell. Stacking both signals produces 2.4 times more citations than directories alone, 5 times more than authority alone, and 5 times more than having neither. Moving from one strong signal to two adds another 28 percentage points of citation rate roughly doubling AI visibility on top of either lever.

Why does stacking compound? AI engines look for confirming evidence across multiple authority surfaces. One signal looks like noise. Two signals look like a pattern. Three signals look like a brand. The more independent surfaces agree on you, the more confidence the model has in surfacing you in an answer.

If you are a SaaS founder reading this, the strategic read is simple. The top-right is the moat. HubSpot, Salesforce, Asana, Trello, Atlassian and Pipedrive live there. The bottom-left is invisible. No single tactic moves you out. Both layers need investment. The two off-diagonal cells are partial wins and the article’s advice lives in the gap between them.

Does directory count or review volume matter more for AI citations?

The number of directories you are listed on barely correlates with AI citations. The number of reviews on those listings does. Directory count is a secondary lever, not the primary one. Review depth is what compounds.

Across the full 137-company sample, directory count correlates with citations at 0.37, a mild relationship that means there is a connection but it is not the main driver. Directory review volume correlates at 0.65, a solid relationship reliable enough to plan a strategy around. The gap between those two numbers is the difference between checking 23 boxes and being talked about on the directories that matter.

Reviews compound. Listings do not. Being on 12 directories with thousands of reviews each beats being on 23 with thin profiles every time, in every category we measured.

The cohort variance tells a tactical story. CRM and Project Management treat reviews as decisive. The relationship inside CRM is so tight that knowing the review volume of a CRM brand tells you almost everything about how often ChatGPT will cite it. Project Management is nearly identical. Dev Tools sits one tier down because community signals like GitHub and HackerNews substitute for directory reviews. HR is half the predictive power because closed RFP-driven buying short-circuits open search behaviour. Marketing Automation and Analytics show only mild relationships because creator endorsements, founder content, and developer mentions drive discovery alongside reviews in those spaces.

Inside CRM specifically there is a clear pattern at the top end. Every brand in our dataset with 10 or more ChatGPT citations had over 25,000 directory reviews: HubSpot at 51,351, Zoho at 34,406, Salesforce at 75,457, Pipedrive at 26,778. This is the incumbent floor for high-volume CRM citation. It is descriptive, not prescriptive. Several CRM brands with substantial review volume still earn zero citations, because review volume on its own is not the lever. Category coherence across signals matters, and so does topical authority on category-defining queries. The 25,000 threshold tells you what the brands AI has been trained to recognise look like, not what you need to do tomorrow.

Do G2 stars or G2 review counts drive ChatGPT citations?

Higher G2 ratings do not produce more ChatGPT citations. Higher review counts do. The data says the opposite of what most SaaS marketers assume.

We pulled the rating data from Blastra’s records and matched it against citation outcomes. Across the 127 companies with G2 rating data, average rating correlates with ChatGPT citations at -0.19. That is a slight inverse relationship, meaning higher G2 stars correlate slightly with fewer citations. The effect is small, but the direction is clear and it is not what most teams expect.

Review count tells a different story. G2 review count correlates with citations at +0.48, a real positive relationship. The difference is enormous in practice. High-volume G2 brands with at least 1,000 reviews average 8.1 ChatGPT citations across our 120 buying queries. Low-volume G2 brands average 2.2 citations, even though their average rating is slightly higher (4.55 stars versus 4.47).

The same pattern shows up on Trustpilot, slightly more pronounced. Trustpilot rating correlates with citations at -0.29, a mild inverse relationship, while Trustpilot review count correlates at +0.32. The conclusion holds across both major review platforms.

This finding is consistent with research G2 itself published in April 2026. Kevin Indig, an advisor to G2, analysed 84,623 products over 180 days and found a similar pattern at much larger scale. Within G2’s free tier alone, going from 0 to 500+ reviews lifts median AI citations from 4 to 3,249, an 812x lift. Within paid, the same progression takes citations from 14 to 6,352, a 454x lift. Reviews correlate with citations at 0.41 for paid profiles and 0.26 for free, comparable in shape to our +0.48 across the merged unified dataset. Indig’s conclusion: “Reviews are the largest single driver.” Two independent datasets, one covering 137 brands across 23 directories and one covering 84,623 G2 profiles, converge on the same answer.

One important confound to name. The inverse correlation between rating and citations (-0.19) likely overlaps with brand size. Big incumbents have broader customer bases, more diverse review pools, and therefore lower average ratings. Small specialty tools have passionate niche users who skew ratings high. So the inverse might be partly “incumbents get cited more” rather than “AI prefers complaint volume.” The volume relationship is independent of that confound and stronger. Reviews drive citations. Stars do not.

Why might this happen? AI engines treat directories as popularity signals rather than quality signals. High volume tells the model the brand is real, in market, and being talked about. High average rating with low volume tells the model the brand is small and possibly cherry-picked. A brand with 5,000 reviews at 4.2 stars produces more comparison content, more web mentions, and more category conversation than a brand with 200 reviews at 4.9 stars. The AI corpus rewards what the web talks about.

The action is uncomfortable for marketing teams used to chasing five-stars. Stop optimising for star ratings. Stop curating only positive reviews. Drive volume across the directories that move the needle. Real customer conversation, even mixed reviews, produces more citation lift than a hand-picked feed of 4.9-star testimonials.

Which directories actually drive AI citations?

Several directories produce a measurable lift in AI citation rate. The rest are noise. We computed the difference in citation rate for each directory by comparing brands listed there to brands not listed there. The hierarchy that emerges does not match the conventional wisdom about which review sites matter most for SaaS. The chart below shows the 13 directories with enough sample size for a reliable comparison.

Crunchbase tops the list, and the reason is structural. Every single SaaS brand in our dataset that was NOT on Crunchbase had zero ChatGPT citations. Every cited brand was on Crunchbase. The sample size on the not-listed group is small, but the pattern is striking. Crunchbase functions as a baseline trust signal that AI engines use to verify the brand exists at all.

TrustRadius beats G2 on lift. The reason is review depth and B2B buyer alignment. TrustRadius reviews are longer, more structured, and read like editorial product evaluations. AI engines treat that quality difference as a stronger signal. Crozdesk and SoftwareReviews score higher than expected because their listings are parsed as editorial-quality directory data rather than raw user-generated content. Both punch above their public profile.

G2 sits at +24 percentage points, which is meaningful but not the headline lift most SaaS teams assume. Being on G2 helps. Driving review volume there helps more. But on a per-brand-listed basis, Crunchbase, TrustRadius, Crozdesk, GetApp and SoftwareReviews all produce a bigger swing in citation rate.

PeerSpot’s +22 lift reflects what makes it different from the other review aggregators. Reviews are long-form and structured, closer in length to enterprise procurement evaluations than to G2’s quick takes, and the platform skews toward dev tools, cybersecurity, and IT infrastructure rather than horizontal SaaS. PeerSpot tends to surface in mid-funnel queries, when buyers move from category awareness to active comparison. AI engines parsing those queries treat PeerSpot listings as evaluation-quality content, which lifts citations on consideration-stage prompts even when the listing volume is modest.

SourceForge produces a +18 percentage-point lift, with the strongest signal in HR and Marketing Automation cohorts. With 132 of 139 brands in our dataset listed there (95% coverage), SourceForge is the closest thing to a universal SaaS directory after Crunchbase. The lift is the listing itself, not the review volume on it, which means SourceForge functions more like Crunchbase as a verification floor than like G2 as a volume engine. Different role, same compounding contribution.

Trustpilot is a special case worth flagging again. It drives a 31-point lift on the volume axis but the rating correlation is slightly negative. AI engines treat Trustpilot as a popularity signal, not a quality source. Get on it. Drive count. Ignore the average score.

Why do some brands with thousands of directory reviews still get zero AI citations?

Pipedrive and Apollo sit in adjacent corners of the sales software market and produce completely different AI outcomes. Looking at them together tells a sharper story than either does alone, because the difference between them is exactly what the Compounding Rule is about.

Pipedrive’s directory layer is unambiguous. Eleven of eleven directories where category matters list Pipedrive as CRM. Its content, its Google rankings and its brand positioning all say CRM. ChatGPT cites Pipedrive consistently for CRM queries. Signals align, AI cites.

Apollo’s directory layer is equally unambiguous in the other direction. Zero of ten directories list Apollo as CRM. Every directory places Apollo in sales engagement, sales intelligence, or lead generation, which matches Apollo’s content authority and brand positioning exactly. ChatGPT cites Apollo for sales engagement queries (it ranked #2 on “best outbound sales prospecting tool” when we tested it) and skips Apollo on CRM queries. Signals align, AI cites.

Same engine. Same week. Two different prompts. On “best outbound sales prospecting tool,” Apollo is the second recommendation, described as an “all-in-one platform: prospect database, email sequences, call tracking, and CRM integration.” On “best CRM alternative to salesforce,” the cited brands are HubSpot, Zoho, Pipedrive, Microsoft Dynamics, Freshsales, Insightly, Copper, Nimble, Keap and Monday.com CRM. Apollo is invisible.

Apollo has 12,597 directory reviews across the directories Blastra crawled. None of that volume moved Apollo into CRM citations, because none of the directories said CRM in the first place.

The pattern is the same in both Pipedrive and Apollo cases. AI engines look for category consistency across third-party signals, and they cite brands where directory placement, topical content, and brand positioning point in the same direction. The Compounding Rule is about whether the signals agree on what category you are in.

That nuance matters because the lazy read of this section would be “directories don’t work.” That is not what the data shows. Directories work when they are saying the same thing as the rest of the web about your category. They get overridden, or more accurately ignored, when an AI engine cannot reconcile what the directories say with what the content says. AI defers to the dominant signal across the whole signal stack. Most of the time, that stack is coherent: directories, topical authority and brand positioning all agree on what a brand is. When they do, AI cites. When they disagree, the dominant signal wins.

The strategic implication is straightforward and it follows directly from the data. AI engines search wide. They pull from multiple source types, directories, editorial content, technical documentation, comparison pages, brand mentions, and they weight them against each other before naming a brand in an answer. You cannot build a viable AI visibility strategy on topical authority alone or directory presence alone. Either lever moves the needle partway. The 84% citation rate belongs to the brands that invest in both, in parallel, and let the signals compound. Single-lever strategies will keep losing to brands that stack.

The pattern is the same in our own client work. Multi-category SaaS brands try to win in two or three adjacent spaces by spreading directory presence across multiple categories. They almost always end up with thin authority everywhere and zero meaningful AI citations in any of them. The Compounding Rule punishes spread. Pick a primary category. Stack both signals there. Win it. Then expand.

How does topical authority and link building feed AI citations?

The brands that win on authority alone have one thing in common. They own their category in content. Not just rankings. Not just backlinks. The combination of category-defining content, category-relevant placements, and category-aligned brand mentions that tells AI engines a single coherent story about what they are.

From the original 150-company AI citation gap study, topical keyword rankings predict ChatGPT citations far more strongly than backlinks alone. The correlation between ranking for a category’s keyword cluster and being cited by ChatGPT comes in at 0.76. Domain rating alone, the metric most agencies still sell on, comes in at 0.34. Ranking for the topical keyword cluster of your category is more than twice as predictive of AI citations as your raw backlink count.

Inside this study, the same pattern shows up even more sharply. Total Google ranking count correlates with citations at 0.29 across all 137 brands but jumps to 0.90 inside CRM and 0.67 inside Project Management. The cohort variance matches what we saw on the directory side. Where buyers research deeply through search, topical authority is decisive. Where buyers rely on community signals or RFPs, the relationship is weaker.

The trend through every layer of this study is the same one: good rankings on Google translate into good rankings on AI. Yes, GEO is different from SEO. Yes, directories matter in ways they did not in classic search. Yes, the sources AI engines pull from include layers no SEO playbook has ever covered. But the dominant pattern in the data is consistent. The brands that earn category-level Google authority are the brands AI engines reach for when buyers ask. Strip away the noise and AI search is mostly classic search done with confirming evidence from third-party surfaces.

Authority-only winners cluster in dev tools. Vercel (49,155 Google rankings, 338 directory reviews → 9 citations). Supabase (73,135 rankings, 54 reviews → 7 citations). Netlify, Render, Linear all follow the same pattern.

How link building fits into the Compounding Rule

Backlinks alone barely move AI citations. The mild correlation of 0.34 tells you that link count is not the lever. This is uncomfortable for the agencies whose pricing model rests on how many DR60+ links they ship per month. The data says counting links is the wrong unit of measurement for AI search.

What moves AI citations is editorial placement on the publications AI engines treat as category-defining. The mechanism is straightforward. When ChatGPT or Google AI Overviews assemble an answer about your category, they pull from a small corpus of authority sources. Get into that corpus and the AI cites you on adjacent queries automatically. That is how the comparison-corpus flywheel works in practice. One placement in the right publication feeds citations across dozens of related buying queries for months.

The way to identify your target publications is simple. Run your top 20 buying queries through ChatGPT and look at the cited URLs. That is your target list. It will be shorter than what most link-building agencies sell you. It will not include random DR60 blogs with “category-relevant” content. It will include five to fifteen publications per category that AI engines actually trust. Win placements there, get named in their next listicle, and the engine inherits the endorsement.

Original research is the single strongest placement type. The AI corpus pulls primary sources first when it can find them. This article is an example. It will become the source ChatGPT pulls from when someone asks about directories and AI citations. Original data with a specific finding compounds harder than any other content format because every secondary discussion of the finding cites it back to you. Publish a study, get it covered, and you are now the primary source for that conversation in the AI corpus for as long as the data holds up.

Comparison content with internal authority anchors comes second. The “X vs Y” page that lives on your own domain, that ranks for the comparison query, and that links into your category content is one of the highest-converting AI citation generators in our client work. Eight of our ten current ChatGPT citation wins are on comparison-style queries. Internal linking that confirms category sits underneath everything, telling AI engines what you do across every page on your site.

Tied together, this is the authority half of the Compounding Rule. Editorial placements in the right publications, original research that becomes the source AI cites, comparison content that owns the buying queries, and internal linking that confirms category. Link building is not separate from topical authority development. Done right, it is topical authority development.

Real client wins, anonymised

We track this in our own portfolio. Live ChatGPT scrapes through our scraping pipeline captured these recent active citation wins across the EMGI client base, all driven by topical authority and editorial placements rather than directory listings.

• A no-code data integration tool ranked third on “best data integration tool for Google Sheets”
• An open-source localisation platform ranked first on “open-source React localisation tool”
• A SaaS FP&A tool ranked first on “best FP&A software for SaaS finance teams”
• An allied health practice management tool featured in the shortlist on “Cliniko vs Halaxy”

Eight of ten current wins across our client base are on comparison-style queries, “X vs Y” or “alternatives to X” or shortlist queries. That is where smaller SaaS brands win in AI search first. The head term (“best CRM software”) corpus is locked up by the both-strong quadrant brands. The comparison corpus has more room because it has more queries and a wider source pool. The tactical workflow we use with every EMGI client goes through this in detail.

For “best project management software for teams,” ChatGPT cited four sources, three listicle pages and one Reddit thread from r/projectmanagers. Reddit drives roughly 40% of AI citation share for SaaS queries. It functions as a credibility booster. Editorial placements and Reddit traction are the same thing in different formats: third-party voices saying you exist and are trusted.

The dev tools exception

Dev tools is the cohort where directories matter least and topical authority matters most. The exception proves the Compounding Rule. The principle still holds. The surfaces are just different.

Vercel, Supabase, Netlify, Render and Linear all sit in the bottom-right quadrant of our 2×2. Low directory presence, high topical authority, still cited. The first thing I did was check what ChatGPT pulled from when it answered dev-tool queries, because if the directory layer was hollow, something else had to be doing the lifting.

When we ran “best deployment platform for nextjs apps” through ChatGPT, the response named Vercel, Netlify, AWS, Render and Cloudflare Pages. The response cited zero directory pages as sources. Not G2. Not Capterra. Not Gartner. The entire answer was constructed from technical documentation, framework blog posts and developer-community content.

The same shape shows up in product analytics. PostHog has barely any presence on the major review directories that lift CRM and PM brands. It still appears in ChatGPT answers for “best product analytics platform for startups,” and it dominates ChatGPT answers for “best open source product analytics” where it ranks first. The question I wanted to answer was: which specific pages are AI engines pulling from when they cite PostHog? We captured the source URLs directly.

This is why the topical authority half of the Compounding Rule still does the heavy lifting in dev tools. If you build content authority on the queries your buyers ask, your name lives in the corpus AI engines parse. That corpus then becomes the answer. PostHog also wins on “PostHog vs Mixpanel” comparison queries, and the reason is structural: PostHog has built its own first-party content authority on the comparison itself. The brand is the query and the answer at the same time.

What AI engines actually use for dev tool queries

• GitHub stars, forks, and issue activity (functions as the dev-tool equivalent of review count)
• HackerNews threads (high-trust developer commentary)
• Reddit programming and framework subreddits
• npm, PyPI and cargo download volume
• Stack Overflow tag activity
• Engineering blog posts on Substack and personal sites
• Editorial category pages like openpanel.dev, thealtstack.com and osssoftware.org for product analytics specifically

These signals serve the same function as directory reviews on the non-dev side of the dataset. They tell AI engines that real users discuss and use this tool. The Compounding Rule still applies. The surfaces are just different. For dev tools, GitHub presence plus topical authority on engineering keywords beats directory listings every time. For non-dev SaaS, directory volume plus topical authority beats either alone. Same framework, different inputs. Stack at least two third-party authority signals AI engines can read, and pick the surfaces that fit your category.

What is the 85% rule for AI citations?

Among SaaS brands listed on Gartner Peer Insights with above-median Google rankings, 85% are cited by ChatGPT. Brands with neither hit 32%. The compounding lift is the article’s strategy in two numbers.

The lift from stacking is +14 percentage points over either signal alone, and +53 over having neither. Gartner specifically because it is the directory AI engines weight highest in the corpus they parse. Gartner’s editorial standards translate into AI trust signals. But the real story is the slope from 32 to 85. Each signal you add doubles your odds of being in the answer. Stack them and you are no longer fighting for visibility. You are the answer.

The bottom-quadrant losers (anonymised)

Some brands have neither directory volume nor topical authority. They are invisible in AI search. We will not name them, the data point is the pattern, not the names. Five archetypes from our dataset, sketched lightly enough that the pattern is visible without the cruelty.

• The open-source analytics tool. One directory listing, 815 Google rankings, zero citations. Strong on GitHub, weak on review-driven discovery. Could win on the dev-tools authority path if it leaned into it.
• The calendar-first PM tool. One directory entry, 12 Google rankings, zero citations. Newer brand with no topical foothold. Either signal alone would be a start.
• The boutique email automation tool. Eight directories, 129 reviews, 126 rankings, zero citations. Listed everywhere, deep nowhere. Textbook breadth-without-depth failure.
• The dev-onboarding analytics tool. Four directories, 41 reviews, 815 rankings, zero citations. Adjacent category positioning issue, the kind that punishes multi-category pitches.
• The new product analytics challenger. Five directories, 14 reviews, zero citations. Bottom-left quadrant resident.

None of these are bad products. All of them are invisible until they invest in both layers. The Compounding Rule is unforgiving on this point. Single-signal investment buys partial visibility. No-signal investment buys the bottom-left.

The playbook: what to do this quarter

The strategy is one sentence. Build directory volume on the five directories that move the needle, and build topical authority on the 50 to 100 buying queries that define your category. Both at once, this quarter, not next.

Directory layer

Audit your current presence. Most SaaS founders are listed everywhere and meaningful nowhere. The lift chart shows 13 directories with measurable AI citation impact. The practical playbook narrows that to a working set of seven you should actively invest in, plus Crunchbase as a non-negotiable verification listing on top.

The seven directories that produce the largest citation lift are TrustRadius, Crozdesk, GetApp, SoftwareReviews, Gartner Peer Insights, Trustpilot, and G2. Each of these produces a citation lift of +24 percentage points or more in our dataset. They cluster around two strengths: editorial-quality long-form reviews (TrustRadius, Crozdesk, SoftwareReviews, Gartner) and high-volume buyer-side review aggregation (GetApp, Trustpilot, G2). Stacking presence and volume across these seven covers both modes.

On top of those seven, Crunchbase is non-negotiable. Every brand in our dataset NOT on Crunchbase had zero ChatGPT citations. This is the cheapest, easiest single directory action you can take and the consequence of not doing it is total invisibility. Confirm your Crunchbase listing exists, is claimed, is current, and describes your category in plain language.

Drive review volume on the seven active investments. Set a quarterly target. The CRM brands in our sample that win at scale tend to sit above 25,000 directory reviews, though this reflects incumbents who already dominate the corpus, not a threshold every challenger must clear before being cited at all. Smaller categories scale proportionally. Skip Trustpilot rating optimisation, drive count not stars. Then stop adding directories. Listing breadth is a secondary lever at best.

Authority layer

Topical content depth on 50 to 100 category-defining buying queries, not 1,000 thin posts. The buying queries that matter are the ones a real prospect types into ChatGPT or Google when they are deciding. Mine your own sales call transcripts for them. Then build comparison content, alternatives content, and category-defining guides.

Editorial link building into the publications AI engines treat as authoritative in your category. The list is shorter than most agencies admit, usually five to fifteen publications. Run your top 20 buying queries through ChatGPT and look at the cited URLs. Those are your targets. Forget DR-based prospect lists. The metric that counts is whether a publication appears in the AI answer for queries you want to own.

Brand mention campaigns across the trade press and editorial round-ups that already rank for those buying queries. Brand mentions carry comparable AI citation weight to direct placements, and they are usually cheaper to earn at volume. Both feed the same authority corpus.

Reddit presence on the subreddits ChatGPT actually pulls from. Roughly 40% of AI citation share for SaaS queries comes through Reddit. That is the third surface of the authority half, and most SaaS teams under-invest in it because Reddit looks like a community channel rather than an authority channel. AI engines do not see the difference.

For dev tools specifically, GitHub and HackerNews replace the directory layer entirely. Same Compounding Rule, different surfaces.

This is the lane EMGI runs. Editorial link building, brand mentions, and Reddit campaigns built specifically for AI citation outcomes, not link counts. If you want to see how the playbook would look for your category, the analysis takes 30 minutes. Email matt@emgigroup.com directly.

The 90-day check

Before you start the campaign, run those top 30 buying queries today and log the result. That is your Day 0 baseline. Without it, you cannot tell whether anything moved.

At 90 days, run the same 30 queries through ChatGPT, Perplexity and Google AI Mode again. Log brand mentions. Compare to baseline. The metric is share of model, how often your brand appears versus competitors in the AI answer. Track quarterly. Adjust per cohort patterns we have laid out.

What this means for SaaS going forward

Single-channel AI visibility strategies are over. The Compounding Rule is the new baseline. The brands winning ChatGPT citations invested in two underlying authority systems for years, and AI engines are now reading both. SEO, reviews, and brand mentions used to be optional optimisations. They are now the discovery layer.

Founders who treat AI visibility as a content sprint will lose to founders who treat it as a stacked authority play. Per category, the precise mix differs. CRM weights review volume highest. Dev tools weight community signals highest. Project Management weights both plus Reddit. The framework is the same. The surfaces vary.

The next 12 months will reward category fit even more heavily, which raises the bar for cross-category brands. Brands that try to be everywhere from day one will keep finding themselves loud on directories and invisible in AI answers. The Compounding Rule is the corrective.

Stacked together, citation rate climbs from 34% to 84%. That is the Compounding Rule. That is the entire playbook.

The 90-second walkthrough (for sharing)

If you want to send this study to someone who will only give you 90 seconds, here are the bullets:

• We merged Blastra’s directory crawl (139 SaaS brands, 23 directories, 2,095 listing checks) with EMGI’s AI citation study (150 SaaS brands, 120 buying queries). 137 companies appear in both datasets, which means we measured the exact same brands on both sides.
• The headline finding is the Compounding Rule. Brands with high directory review volume and high topical Google authority get cited by ChatGPT 84% of the time. Brands with neither sit at 34%. Either signal alone gets you partway. Stacked, they win.
• AI engines converge. Google AI Overviews and ChatGPT cite the same SaaS brands more than 9 times out of 10 (correlation 0.91). We verified the same convergence on Perplexity Sonar and Gemini 2.5 Flash. The signal layer is shared.
• Reviews beat ratings. G2 average rating correlates with AI citations at -0.19 (slight inverse). G2 review count correlates at +0.48. AI rewards conversation volume, not satisfaction scores.
• Not all directories matter equally. Crunchbase, TrustRadius, Crozdesk, GetApp, and SoftwareReviews produce the largest citation lift. G2 still matters but is not the top performer it is often assumed to be.
• Apollo has 12,597 directory reviews and zero ChatGPT citations on CRM queries. Same brand ranks #2 on “best outbound sales prospecting tool.” Directories alone are not strong enough. AI defers to the dominant signal, and the dominant signal is the web.
• Dev tools is the exception that proves the rule. Vercel, Supabase, PostHog and Linear win on authority alone because AI pulls from GitHub, Reddit, and editorial category pages instead of review directories. Same Compounding Rule, different surfaces.
• Trend through every layer of the data: good Google rankings still equal good AI rankings. GEO is SEO with confirming evidence from a second authority surface.
• The playbook is one sentence. Build directory review volume on Gartner, TrustRadius, Crozdesk, SoftwareReviews and G2. Build topical authority on 50 to 100 category-defining buying queries. Both at once. Quarterly check on share of model.

Frequently Asked Questions

Do I need to be on every directory?

No. The number of directories you are listed on barely correlates with AI citations (mild relationship, 0.37). The number of reviews on those listings does (solid relationship, 0.65). Concentrate effort on the seven directories with the largest measured citation lift (TrustRadius, Crozdesk, GetApp, SoftwareReviews, Gartner Peer Insights, Trustpilot, G2) and confirm Crunchbase listing as the verification floor. Listing breadth is a secondary lever, not the primary one.

Does my G2 average rating matter for ChatGPT citations?

Less than you think. G2 average rating correlates slightly negatively with ChatGPT citations (-0.19). G2 review count correlates positively at +0.48. AI engines reward conversation volume, not satisfaction scores. Chase reviews, not stars. A brand with 5,000 reviews at 4.2 stars produces more citation lift than a brand with 200 reviews at 4.9 stars.

Why is Apollo not cited by ChatGPT on CRM queries?

Apollo is consistently categorised as sales engagement, sales intelligence or lead generation across every directory that lists it (0 of 10 directories place Apollo in CRM). Apollo’s content authority, Google rankings and brand positioning all say the same thing. ChatGPT cites Apollo for sales engagement queries, where it ranks #2 on “best outbound sales prospecting tool,” and skips Apollo on CRM queries. Signals agree, AI cites accordingly. The Compounding Rule rewards category consistency, not just review volume.

Does Trustpilot help SaaS get cited by ChatGPT?

Yes, but only the review count matters. Trustpilot review volume drives a 31-point citation lift. Average rating shows a slight inverse correlation (-0.29). AI engines treat Trustpilot as a popularity signal, not a quality source. Get on it, drive count, ignore the average score. The same pattern shows up on G2 just slightly less pronounced.

Do Google AI Overviews and ChatGPT cite the same brands?

Yes, almost in lockstep. Across 137 SaaS companies and 120 buying queries, the two surfaces correlate at 0.91. We verified the same convergence on Perplexity Sonar and Gemini 2.5 Flash. Three engines, same brand cluster.

What is the fastest way to get cited by ChatGPT for SaaS?

Stack two third-party authority signals at once: high directory review volume on Gartner and TrustRadius, plus topical Google authority on 50 to 100 category-defining buying queries. The compounding effect produces an 85% citation rate compared to 32% for brands with neither. For dev tools, replace directories with GitHub, Reddit and HackerNews presence.

Conclusion

The Compounding Rule is the entire argument. Directory volume and topical authority are two pillars, not one foundation. Neither alone is enough. Together they compound, and the 84% citation rate belongs to the brands that invest in both. 51 brands occupy the both-strong quadrant. They average 8.2 ChatGPT citations across 120 buying queries. Brands with neither average 1.6. The gap between those numbers is the strategy.

The pattern holds across ChatGPT, Google AI Overviews, Perplexity Sonar and Gemini 2.5 Flash. This is how AI search works across every surface we tested.

If you want to see where your SaaS sits in the quadrant, drop me a line at matt@emgigroup.com and I will run the analysis on your category personally.