How to Get Your SaaS Product Recommended by ChatGPT, Gemini and Perplexity

The five SaaS prompts that decide your shortlist

For a SaaS product, "recommended" means being named as a fit, in a particular order, for one buyer intent. A mention in a paragraph about the category is weaker. A citation of your blog post that doesn't name you as an option is weaker still. The distinctions are covered in AI mentions, citations, and recommendations.

Buyer prompts in SaaS cluster into five types. Each type tends to be answered from different evidence, so each needs a different asset.

The "what usually backs the answer" column is our editorial synthesis. It is not a measured split. One data point supports it directly: in Parse's data, 21.4% of cited vendor pages carried roundup-style titles such as "best", "top N", "versus" or "comparison" (Parse). Engines are already treating vendor comparison content as evidence for shortlists.

Treat each prompt type as a family of phrasings, not a single query. Answers vary between identical reruns, and they vary more once the wording changes. Adding a constraint ("for a 20-person team", "that integrates with Salesforce") is the kind of change most likely to reshuffle the list. You win a prompt type by owning the evidence behind many phrasings of it, not by tuning for one phrasing.

Takeaway: pick the two or three prompt types you should win on merit. We'd start with head-to-head and constraint prompts, because your own pages carry weight there. A broad "best [category]" prompt in a crowded market is likely to be the last to move.

Why AI recommends your competitor instead of you

A competitor gets recommended over you because its evidence is easier for the engine to find and corroborate on that prompt. Ranking on Google does not settle this. A page can rank #3 in Google and still be absent from ChatGPT's shortlist, a pattern examined in Why does a page rank in Google but still get ignored by AI?.

Diagnose before you produce content. We use four states. This is our working framework, not a published taxonomy. Each state has a different fix.

To tell the states apart, run two prompts side by side. Use one unbranded prompt ("best onboarding software for PLG startups"). Use one branded capability prompt ("does [you] support in-app checklists?"). If you fail both, you are in state one. If you pass the branded prompt and fail the unbranded one, you are in state two. If you appear only when the prompt adds a segment or constraint, you are in state three.

The most useful single test is to open the sources cited when the competitor is recommended. Note who published each one: the competitor's own site, an independent publisher, a community thread, or a directory. That list is your roadmap. If ChatGPT is citing the competitor's own comparison page, you need a better one. If AI Mode is citing a roundup you are missing from, you need that placement.

Run the diagnosis on every engine separately. A competitor that dominates ChatGPT may be beatable on AI Mode, where the evidence mix is different.

Then check whether the category is already decided. A category is decided when one competitor leads on every engine for every phrasing. If that is your situation, you are doing displacement work, and it takes longer. The practical response is a narrower wedge: pick the segment or constraint where you are the better fit, and win those prompts first.

Sometimes the diagnosis shows the problem is positioning, not visibility. If engines describe you accurately and still pick a competitor for your target segment, they may be right about fit. More content will not fix that.

ChatGPT, Gemini, Perplexity and Google AI don't build shortlists from the same sources

Each engine has its own eligibility control and draws on a different mix of evidence, so the same audit leads to different fixes. The public data is uneven. Parse's large study covers only ChatGPT Search and Google AI Mode. The inspected studies do not break out source mixes for the Gemini app or Perplexity.

ChatGPT

ChatGPT search inclusion is controlled by OAI-SearchBot. OpenAI says sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers," though they can still appear as navigational links (OpenAI).

The other OpenAI crawlers do different jobs:

• GPTBot governs training. Its setting is independent of OAI-SearchBot, so blocking GPTBot does not remove you from ChatGPT search.
• ChatGPT-User fetches pages when a user asks. OpenAI says it is not used to decide whether content may appear in Search.

ChatGPT leans on vendors' own pages more than Google does. Parse found that the recommended brand's own site supplied 50.4% of recommendation-backing citations on ChatGPT Search (Parse). Separately, across Parse's whole dataset, commercial media supplied just 11.0% of recommendation evidence in the Software industry. That is an industry average across both engines Parse studied, not a ChatGPT-only share. The data shows correlation, not cause, and 13.2% of citations could not be classified. We break down the ChatGPT-specific source audit in How to get mentioned by ChatGPT.

ChatGPT's source mix is also moving. AI-visibility trackers reported during 2026 that ChatGPT's citations shifted further toward company-owned pages and away from Reddit and forums. Other engines did not show the same shift. The size and cause of that change could not be confirmed from primary data. Check your own citation logs before reprioritizing around it.

Whether Bing rankings predict ChatGPT recommendations is unsettled. No inspected evidence tests it, so do not use Bing position as a proxy.

Implication: on ChatGPT, your own comparison, pricing, docs and help pages are the first lever.

Google AI Mode and AI Overviews

Google says AI Overviews and AI Mode have no extra requirements and need no special optimization. A page must be indexed, eligible to show a snippet, and meet Search's technical requirements. Google also says you do not need AI text files, new machine-readable files, or special schema.org markup (Google Search Central).

Google adds that AI Mode and AI Overviews may use query fan-out and different models. That means the two surfaces can cite different links for the same question.

AI Mode draws on a different evidence base than ChatGPT. In Parse's data, community and social sources supplied 14.3% of AI Mode's recommendation-backing citations, against 5.2% on ChatGPT Search. Independent publishers supplied 37.6% on AI Mode, against 29.4% on ChatGPT (Parse).

Peec found smaller listicle-rank associations on AI Overviews and AI Mode than on engines that cite fewer sources, such as ChatGPT (Peec AI).

Implication: on Google's surfaces, classic SEO and independent coverage matter more, and your own pages carry proportionally less weight.

Gemini

The Gemini app is not governed the way Search is. Google points to Google-Extended as the control for limiting AI training and grounding in its other systems (Google Search Central). Blocking Google-Extended can affect Gemini grounding while leaving AI Mode and AI Overviews untouched, because those depend on Search indexing.

Public, SaaS-specific data on which sources Gemini cites is thin in the evidence we reviewed. Measure Gemini as its own engine rather than assuming it behaves like AI Mode.

Perplexity

Perplexity inclusion is controlled by PerplexityBot. Perplexity says PerplexityBot surfaces and links websites in its answers and is "not used to crawl content for AI foundation models" (Perplexity). Allowing it therefore has no training trade-off.

Perplexity publishes IP ranges and firewall allow-rule steps for Cloudflare and AWS WAF. Changes can take up to 24 hours to apply. Perplexity-User, which fetches pages for individual queries, generally ignores robots.txt.

The inspected studies give no Perplexity source-mix breakdown. To rank in Perplexity for your category, record what it cites for your prompts and treat that as the evidence set.

Make your site eligible on all five AI surfaces

Eligibility is the precondition, not the lever. If any engine cannot reach your pages, no amount of content helps on that engine. The checklist:

• ChatGPT: OAI-SearchBot is allowed in robots.txt, and OpenAI's published IP ranges are allowed at your firewall or CDN. robots.txt changes take about 24 hours (OpenAI).
• Perplexity: PerplexityBot is allowed in robots.txt and in your WAF rules.
• Google AI Mode and AI Overviews: money pages are indexed in Google and snippet-eligible. Check that no nosnippet directive sits on pricing or comparison pages. Google also notes that crawling must be allowed at the CDN and hosting level.
• Gemini app: make a deliberate decision on Google-Extended, knowing it affects grounding rather than Search.
• Rendering: key product copy should be present in the served HTML. This is a judgment call, not a documented requirement. If a crawler does not execute JavaScript, it will not see copy that is injected client-side.

The full OpenAI crawler and firewall setup is in How to get mentioned by ChatGPT.

The common failure is silent. A WAF rule blocks OAI-SearchBot, and nobody notices until the brand is missing from ChatGPT. We make Tideflow AI, and our AI bot detection runs server-side through a Cloudflare route Worker (recommended) or Express middleware. It shows whether OAI-SearchBot, PerplexityBot, GPTBot, ChatGPT-User and others actually fetched a page. We mark a visit as verified only when Cloudflare's bot-management metadata confirms it. A crawl is a discovery signal. It does not prove indexing, citation or recommendation.

Build the owned SaaS pages engines use as evidence

For B2B SaaS, your own site is a legitimate and heavily used evidence source, especially on ChatGPT. Parse found own-site shares of 40.2% in Professional Services, 37.1% in Data and Analytics and 31.7% in AI, while commercial media stayed near 11–13% in Software, IT and Sales and Marketing (Parse). These are industry averages. Your own audit of cited sources should override them for your category.

Comparison and alternatives pages

An honest "[you] vs [competitor]" page is the most direct way to win head-to-head prompts. An honest "[competitor] alternatives" page does the same for alternatives prompts. "Honest" has an operational meaning here:

• Concede the use cases where the competitor is the better choice.
• Link to public sources for claims about the competitor.
• Put a "checked [month, year]" date next to any pricing.

A page like this is more credible to readers and harder to dismiss as self-promotion. If you publish your own roundup, disclose that you make one of the listed products, and rank on stated criteria. The risks of self-ranked and paid lists are covered in Are third-party listicles good GEO or just AI-era link building?.

An illustrative weak page: "Why [you] beats [competitor]" is a feature grid where every row is a checkmark for you.

An illustrative strong page: "[You] vs [competitor]" names the three buyer segments each tool fits. It says plainly that the competitor has a deeper Salesforce integration. It then shows where you win on setup time and pricing, with the date the pricing was checked.

Use-case and integration pages

If you're a smaller SaaS brand, we'd start with constraint prompts. A prompt such as "[category] that works with HubSpot for a 20-person team" narrows the field. The engine then needs evidence that a specific product fits that exact constraint.

Build one page per segment you actually serve: team size, industry, or stack. Build one page per major integration. Each page should state the constraint in the buyer's words ("for agencies managing 10+ clients"). It should also explain concretely how the product handles it, with setup steps or plan requirements. Do not create pages for segments you serve poorly. Engines will eventually corroborate you against reviews and third-party coverage, and so will buyers.

Pricing, docs and help centre

Validation prompts are answered with facts, so publish the facts. "Is [you] worth it for [use case]" and "[you] pricing" are late-stage prompts. They need a public pricing page with plan limits, crawlable docs, and clear security and compliance statements.

If your pricing sits behind a demo form, the engine will fill the gap from whatever third party last guessed at it. That is how you end up in the "described wrong" state.

One positioning sentence, used everywhere

State your category, audience and problem the same way on your homepage, directory profiles, partner and marketplace listings, founder bios and press boilerplate. This is practitioner judgment, not a measured effect. Inconsistent descriptions make it harder for an engine to decide which prompts you belong in.

A workable template: "[Product] is a [category] for [audience] that helps them [solve problem]."

How to write the passages inside these pages so they get cited is covered in What content do AI models cite most?.

Our content creation drafts these pages from the gaps our monitoring finds. It then publishes approved drafts through MCP to your own AI coding agent and website stack. Nothing goes live without human review.

Get into the third-party sources each engine cites for your category

Target the few third-party sources engines already cite for your prompts, not every list on the web. Evidence in AI recommendations is widely dispersed. In Parse's median category, the top domain supplied only 7.4% of evidence, drawn from 260 domains in total (Parse). No single site is a gate. That makes it important to know which sites are actually cited for your specific prompts.

Cited listicles first

Inclusion in a listicle that engines repeatedly cite is associated with being recommended, and a higher position in it goes with an earlier position in the answer. In Peec's observational study of about 200,000 responses across eight engines, B2B SaaS brands ranked #1 in such a listicle showed +16.5 percentage points of visibility and were named 1.17 positions earlier (Peec AI).

Two qualifications matter:

• Returns diminished after a few repeated placements.
• In established B2B SaaS, inclusion alone delivered most of the lift.

Peec sells visibility tracking, and it cautions against treating any rank tier as a rule. We weigh earned, paid and self-ranked placements in our analysis of third-party listicles.

To find the right targets, pull the roundups cited for your prompts on each engine. Then pitch the publisher for inclusion with evidence they can check: your comparison page, docs, public pricing, and the segment you fit best. A pitch that says "we're the best fit for teams under 50 on HubSpot, here's the integration doc" gives an editor something to verify.

Review directories: corroboration, not a gate

You do not need a review-count threshold on G2 or Capterra to be recommended. Parse's directory study covered 835 categories and 5.4 million evidence links. G2, Capterra, GetApp, Software Advice, SourceForge, TrustRadius and Trustpilot were each the #1 evidence source in zero categories (Parse).

The software-specific figures point the same way:

• Only 24.0% of software-category brands had any directory citation.
• Directories supplied 0.74% of software recommendation evidence.
• Across all categories, directories and review sites supplied 2.9% of recommendation-backing citations (Parse).

Parse's own framing is that directory work is not useless, only not a prerequisite. Keep your profiles accurate, current and reviewed, because they are cited sometimes and buyers read them. Don't sequence the rest of the program behind them.

Communities, YouTube and Reddit: check per engine

Community sources are powerful on some engines and in some categories, not everywhere. In Parse's data, YouTube was the #1 source in 169 categories and Reddit in 142 (Parse). Most of that community weight shows up on Google AI Mode (14.3% of evidence) rather than ChatGPT Search (5.2%).

Participate genuinely in the threads and channels where your buyers ask for recommendations, because that still feeds Google's surfaces. Then check the effect on each engine separately instead of assuming Reddit work moves ChatGPT.

Fix what cited sources say about you

If a cited partner page, marketplace listing or directory profile describes an old plan, an old category or a discontinued feature, the engine may repeat it. We'd correct the source before trying to outrank it. Start with the stale descriptions that appear in answers to your validation and pricing prompts.

Our gap plans turn this audit into specific actions. For each gap on each surface, we show which sources the answer relies on and which competitors appear. The plan then proposes improving an existing page, creating new content, winning the citation, or getting included in sources AI already cites.

Measure recommendation share, rank and signups per engine

Measure each engine separately, over repeated runs of several phrasings per prompt type, and tie the result to signups by landing page. A single prompt run on a single engine tells you almost nothing, because answers vary between reruns and paraphrases.

For each prompt family, write three or four phrasings, including at least one constraint variant. Then track four things per engine:

• Named or not: whether you appear in the shortlist at all.
• Position: where you appear. Peec's data shows mention rate and answer position move somewhat independently, so keep them as two metrics.
• Share of voice: measured against the competitors the engines actually name, not your internal competitor list.
• Own-domain citation: whether your site is among the cited sources.

Do not blend engines into one score. ChatGPT's reported 2026 source shift did not appear on the other engines, and a blended number would have hidden it.

Tie visibility to revenue through landing pages, and be honest about what you can't see. ChatGPT and Perplexity referrals can be identified by referrer domain (chatgpt.com, perplexity.ai), so you can count signups from sessions that start on your comparison or pricing pages.

Google's AI Overviews and AI Mode clicks are counted inside Search Console's ordinary "Web" search type (Google Search Central). They cannot be separated as AI referrals. For Google's surfaces, attribution is inferential: watch landing-page conversions and Search Console trends for the pages you built for AI prompts.

Keep three signals apart:

1. A crawler visit means discovery.
2. An appearance in an answer means inclusion.
3. A referral that converts means business impact.

How to re-test after publishing, and what to look for, is covered in three signals that show ChatGPT picked your pages up.

We run this measurement daily on every plan across ChatGPT, Gemini, Perplexity, Google AI Mode and Google AI Overview. For each prompt we record whether you were mentioned, whether you were cited, your rank and which competitors appeared. Our rank is first-mention order, a measure of prominence, not endorsement. Our analytics aggregate AI referrals and conversions such as signups by landing page. AI Mode and AI Overviews visits arrive as Google search traffic, and we don't pretend to split them out.

Where Tideflow AI fits

Tideflow AI is built to run the loop this guide describes in one place:

1. Measure recommendation share and rank on five surfaces.
2. Use the gap plan to see which page or source to fix.
3. Draft the comparison, alternatives or use-case page.
4. Publish it through MCP into your own stack.
5. Confirm AI bots fetched it.
6. Watch AI referrals and signups by landing page.

Every plan starts with a 7-day free trial, no credit card required, covering 1 brand and 10 prompts. Paid plans, per our plan comparison:

• Starter: $99/month for 1 brand and 50 prompts.
• Growth: $199/month for 3 brands and 100 prompts.
• Scale: $499/month for 10 brands and 250 prompts.

All plans monitor all five surfaces daily. Prices checked September 2026.

Our limits:

• We don't score sentiment.
• We don't track Claude, Copilot or engines beyond the five listed.
• We don't track ChatGPT Ads or Shopping.
• We deliberately don't produce a single blended visibility score.
• We can't guarantee recommendations or rankings.

Profound, Peec AI and Otterly AI are the other serious options. Profound can find gaps, draft content and stage it to your CMS, and it covers more engines on higher tiers. Peec goes deep on citation-source and listicle analysis. Otterly offers broader, lower-cost monitoring that includes Claude. Details are in Best AI visibility tools in 2026 and our comparisons with Profound, Peec AI and Otterly AI.

What won't get your SaaS recommended

• Schema as a recommendation lever. Google says no special schema.org markup is needed for its AI features. Keep valid markup for normal SEO, but don't expect it to move shortlists.
• llms.txt and .md copies of pages. Google says AI text files aren't needed for its surfaces. No inspected evidence shows they move ChatGPT or Perplexity recommendations.
• Review-count thresholds. Parse's 835-category data gives no support for a "get N reviews first" rule.
• Product feeds and shopping programmes. These are built for physical-product shopping, not SaaS shortlists.
• Ad spend. OpenAI documents a separate crawler, OAI-AdsBot, that only visits submitted ad landing pages. Organic search inclusion runs through OAI-SearchBot (OpenAI). We don't track ads.

Frequently Asked Questions

Why does ChatGPT recommend my competitor instead of me?

ChatGPT recommends a competitor when that competitor's evidence is easier to retrieve and corroborate for the prompt. Usually that means its own comparison pages and the roundups ChatGPT cites. On ChatGPT Search, recommended brands' own sites supplied 50.4% of recommendation-backing citations (Parse). Open the sources cited next to the competitor's recommendation. Those pages are what you need to match or join.

I rank #3 on Google. Why doesn't ChatGPT recommend us?

Google rank measures something different from ChatGPT's shortlist evidence. ChatGPT leans more heavily on vendors' own comparison, pricing and docs pages than Google AI Mode does. It also weights cited listicles differently. Whether Bing rankings predict ChatGPT results is unsettled. Audit what ChatGPT cites for your prompts rather than inferring from Google position.

Can I submit or list my product in ChatGPT?

OpenAI's crawler documentation describes no submission path for organic answers. Any public site can appear in ChatGPT search if OAI-SearchBot is allowed to reach it through both robots.txt and your firewall. Once you are eligible, whether you get recommended depends on the evidence ChatGPT finds on your site and in the sources it cites.

Do I need G2 or Capterra reviews to get recommended?

No. Parse found G2, Capterra and other major directories were the top evidence source in 0 of 835 categories. Directories supplied just 0.74% of recommendation evidence in software categories. Keep your profiles accurate and current, because they are occasionally cited and buyers read them, but don't treat them as a prerequisite.

Does posting on Reddit still help get recommended?

It depends on the engine. Community and social sources supplied 14.3% of Google AI Mode's recommendation evidence against 5.2% on ChatGPT Search (Parse). Trackers also reported ChatGPT's Reddit citations falling in 2026, for reasons that haven't been confirmed. Participate where your buyers genuinely ask for advice, then measure the effect on each engine separately.

Do Gemini and Perplexity recommend the same tools as ChatGPT?

Often not, because each engine draws on a different mix of sources and has its own eligibility control. Public SaaS-specific data on Gemini and Perplexity source mixes is thin. Track each engine separately with the same prompt set, and diagnose gaps per engine rather than assuming one engine's result carries over.

How long until ChatGPT starts recommending us?

robots.txt changes take about 24 hours to reach OpenAI's crawlers, and up to 24 hours for Perplexity. Google recrawling can take longer. No reliable timeline exists for new content turning into recommendations, because that depends on what each engine retrieves and cites. Re-test on a fixed schedule and watch whether your pages start appearing among cited sources.

Your first move this week

1. Pick your three highest-value prompt types.
2. Write three or four phrasings of each, including one constraint variant.
3. Run every phrasing on ChatGPT, Gemini, Perplexity, Google AI Mode and AI Overviews.
4. For each prompt and engine, log who is recommended, in what order, and which sources are cited.
5. Fix the single largest source gap first. That is usually a missing comparison page for ChatGPT, or a missing roundup placement for AI Mode.

If you'd rather not run that baseline by hand, a 7-day trial with no card runs it daily across all five surfaces: Start free.

Sources

• Parse, Who publishes the pages AI cites for buying advice?; The directory tax is a myth: G2 and Capterra are #1 in 0 of 835 categories
• OpenAI, Overview of OpenAI Crawlers
• Google Search Central, AI features and your website (December 2025)
• Perplexity, Perplexity Crawlers
• Peec AI, The Listicle Rank Effect: What 200K AI Responses Across 8 AI Engines Reveal About Brand Visibility (May 2026)