We asked OpenAI's model with web search and Perplexity ten questions a help desk software buyer might ask, like "best help desk software for small business" and "Zendesk alternatives." Help Scout came up in 15 of the 20 answers. HappyFox, which sells help desk software at similar prices and has been around since about 2011, came up in none.
That gap is the whole game when you're trying to improve your AI search rankings. So we went looking for its cause, checked the obvious suspects, and found the difference somewhere most SaaS teams don't look.
What the engines cited
The 20 answers carried 263 citations between them. Here's how they sorted by page type:
| Page type | Citations | Share |
|---|---|---|
| Third-party roundups and "best of" posts | 109 | 41% |
| Vendor product pages | 60 | 23% |
| Vendor pricing pages | 53 | 20% |
| Vendors' own comparison pages | 22 | 8% |
| Review sites like G2 and Capterra | 19 | 7% |
| Reddit, YouTube and forums | 0 | 0% |
The two engines read differently. Perplexity's 190 citations leaned on roundups, 94 of them, while 46 of OpenAI's 73 citations were vendors' own pricing pages. Across both, pages written by someone other than the vendor carried nearly half the weight. Reddit got plenty of name-checks inside the answers, but not one direct link.
Help Scout has one built-in advantage worth admitting: one of our ten questions was "Help Scout vs Zendesk." Leave that one out and it still showed up in 13 of the other 18 answers.
What didn't explain the gap
Content quality wasn't it. HappyFox's comparison page against Zendesk runs about 3,700 words, a little longer than Help Scout's, and both companies publish their prices in plain HTML that any crawler can read without running JavaScript.
The robots.txt looked like a smoking gun at first. HappyFox's file blocks GPTBot, ClaudeBot, Bytespider and Common Crawl's CCBot outright, while Help Scout's blocks nothing. Read it again, though, and it's a deliberate, consistent policy. The same file carries a content signal of search=yes, ai-input=yes, ai-train=no, and the four blocked bots are training crawlers. ChatGPT's search runs on a different crawler, OAI-SearchBot, and HappyFox leaves that one open, along with PerplexityBot.
So ChatGPT's search could read HappyFox's pages. Blocking the training crawlers may mean the models remember less about HappyFox, which matters when an answer leans on memory, but we can't prove that from the outside. Google, for its part, says "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary," so there's no hidden technical switch either.
What did explain it: how often the reading list names you
We call the third-party pages an engine keeps citing for a category its reading list. For help desk software, it ran from Forbes Advisor, PCMag and Capterra down to dozens of vendor blogs and small review sites. We fetched all 104 pages the engines cited and managed to read 91.
HappyFox wasn't missing from the big ones. PCMag's IT help desk roundup ranks HappyFox Service Desk fourth, as its pick for beginners, and Forbes Advisor lists HappyFox as the best option for a multilingual agent portal. The engines cited both pages and still never named HappyFox.
The difference was breadth. Help Scout was named on 57 of the 91 pages, and HappyFox on 21. Most of the long tail left HappyFox out entirely: the "Zendesk alternatives" posts on other vendors' blogs, the small-business roundups and the shared-inbox guides.
We'd call that number your consensus count: how many pages on your category's reading list name you. In our test, a high consensus count beat a few prestigious placements.
We can't prove it causes the gap, since better-known brands get written about more. It's still the lever you can pull this quarter, and it matches what one indie founder found after asking three engines for recommendations in 15 software categories: the models mostly repeat what review sites, comparison pages and Reddit threads say.
200+ SaaS teams already track their AI citations.
They know exactly when ChatGPT mentions their brand, and when it stops. Do you?
How to improve your AI search rankings, step by step
1. Build your category's reading list
Write down 15 to 20 questions your buyers ask before they buy, mixing "best X for Y," "X vs Y," "X alternatives" and a pricing question. Ask each one in ChatGPT and Perplexity with search on, then collect every third-party URL they cite. Mark which pages name you and which name your top two competitors. That list, sorted by how often each page gets cited, is your target list. Expect two to three hours for the first pass.
Here's the top of the help desk list from our test, with the two brands checked:
| Page | Times cited | Names Help Scout | Names HappyFox |
|---|---|---|---|
| PCMag's help desk picks (two URLs, one page) | 6 | No | Yes |
| Forbes Advisor's best help desk software | 2 | Yes | Yes |
| Capterra's help desk category | 2 | Yes | Yes |
| A help desk comparison on CloudTalk's blog | 2 | Yes | Yes |
| A help desk roundup on Drag's blog | 2 | Yes | No |
The top of a list like this tells you where the prestige is. The long tail below it, where each page is cited once, is where the consensus count gets decided.
2. Get onto the pages that leave you out
Aim for count over prestige. Look at who wrote the roundups on your list. In help desk software, many came from other vendors' blogs, including monday.com, HelpCrunch, Capacity and Kustomer. Email the author or editor with your pricing, the one use case you win, and a current screenshot. Update your G2 and Capterra profiles too, since review sites made up 7% of citations on their own.
3. Publish your own comparison pages
Perplexity cited Help Scout's own comparison pages five times, including its "Help Scout vs Zendesk" page, which walks through features, pricing, ease of use and integrations in separate sections.
Write one for each competitor your buyers weigh you against, plus an "alternatives to [your product]" page before a rival writes it for you. Our guide to comparison pages that get cited covers the structure.
4. Keep prices and plans in plain HTML
On OpenAI's side, Help Scout's pricing pages were cited 11 times, second only to Zendesk's. Both companies in our test publish prices in plain HTML, so it's table stakes, not an edge. If your prices hide behind a sales form or load with a script, the engine builds its answer from someone else's description of you.
5. Make your crawler policy a decision
Let the search crawlers in: OAI-SearchBot, PerplexityBot and Claude-SearchBot. Then decide on the training crawlers on purpose, knowing that blocking GPTBot and ClaudeBot keeps your pages out of future model training and may leave the models remembering less about you. Our AI crawler list explains what each bot does.
6. Keep the pages that matter current
Fresh pages get picked more. Ahrefs looked at 17 million AI citations and found the average cited URL was 1,064 days old, against 1,432 days for pages in Google's organic results, with ChatGPT's citations the freshest of any engine. Stale answers also hurt in the other direction, as one marketer found when ChatGPT described a product their company had killed in 2022.
chatgpts answer about our brand was from 2022 and leadership still wants an ai visibility plan this week
asked chatgpt what our brand does and got a confident paragraph about a product we killed in 2022. screenshotted it. leadership saw it and now wants an "ai visibility plan" on the calendar this week we already have ranking reports, social l...
How often does ChatGPT mention your brand?
Most founders have no idea. The answer might surprise you.
What won't move your AI rankings
Some popular fixes didn't show up in our test at all. Google's own guidance says you "don't need to create new machine readable files, AI text files, or markup to appear in these features," and there's "no special schema.org structured data that you need to add." Claims that schema doubles or triples AI citations trace back to agency blogs without a published method.
Reddit is the other trap. It came up inside answers as something buyers read, but not one of the 263 citations was a direct Reddit link. Seeding threads for a link won't get you cited, while showing up in the roundups that people on Reddit point to might. If you're weighing an llms.txt file, our llms.txt review covers what it does and doesn't change.
Measure the same questions every week
None of this shows up in a single check. Answers change from run to run, and a roundup you got added to may take weeks to be re-crawled. Keep your question list fixed and record, every week, whether you were named, whether you were cited, and which pages were cited instead.
By hand, that's a few hours a week once the list passes 20 questions and two engines. Our AEO audit walkthrough shows the manual version on a real SaaS site.
Or RankControl can run it. AI visibility tracking asks up to 50 buyer questions every week on seven engines and shows which competitors got named instead of you and which sources each engine quoted, which is your reading list, refreshed weekly. The content engine then writes the comparison and alternatives pages from those questions and publishes them on your site.
For a first look at where you stand, run a free check of your brand in AI answers.
15 hours a week manually. Or 15 minutes with RankControl.
Track citations, monitor competitors, and fix content gaps across every AI search engine. Automatically.
