Sauror AIGEO for SaaS: the 7 pages ChatGPT reads before it recommends you
When a buyer asks "which invoicing software for my small company?", ChatGPT builds a shortlist, then checks specific facts on each vendor's site: price, free plan, compliance. If your site does not answer, it takes the answer from a competitor or a review site. Here are the 7 pages a SaaS needs, what goes in each one, and how to check in 2 minutes that it works.
What ChatGPT actually does when asked for software
Before answering, ChatGPT writes its own web searches. Across 53 answers collected for buyer questions about French invoicing and accounting software (RobZ IA barometer, August 2026), it wrote 72 of them. Two examples, word for word: "site:indy.fr SASU comptabilité prix 2026" and "site:abby.fr plateforme agréée facturation électronique 2026".
Almost half of these searches (34 of 72) target a single vendor's site. ChatGPT has already picked brands. It comes to check. Of those 34 checks, 13 are about price and 11 about compliance. And 94% of all its searches include the current year.
What it means for you: your site is not a shop window, it is an answer base. One question, one URL, the answer at the top of the page.
Page 1: the pricing page, in text, not in a widget
Why. It is the first thing ChatGPT checks. Three pricing pages (Abby, Pennylane, Indy) collect 15 citations between them. When a review site gives no price, ChatGPT immediately runs a "site:indy.fr prix" search.
The classic mistake. A price shown by a JavaScript configurator or a monthly / yearly toggle. An AI that reads the page without running the code sees empty boxes. It then takes the price from a review site, often out of date.
2-minute test. In a terminal: curl -s https://yoursite.com/pricing | grep $. If your prices do not show up, the AI does not see them either.
What goes in it:
| Element | Example |
|---|---|
| Title with the year | [Product] pricing 2026 |
| One line per plan | Starter: $19 / month billed yearly, $24 billed monthly |
| Limits in numbers | 3 users, 500 contacts, 1,000 emails a month |
| What makes you upgrade | "Move to Pro above 3 users" |
| The price date, visible | Prices as of 1 October 2026 |
| 4 to 6 questions under the table | Is there a commitment? Is the price per user? |
Page 2: the free plan or trial, with the limits written down
Why. "Free X software" is one of the most asked questions. ChatGPT checks on your site what "free" means for you. A vague line ("try it free") gives it nothing to quote.
What goes in it. A dedicated URL (/free or /free-trial). First sentence: what is free, for how long, under what condition. Then a free versus paid table, limit by limit. Then the exact moment the user has to pay.
A quotable first sentence: "[Product]'s free plan lets you send 10 invoices a month, with no time limit and no credit card."
Page 3: "[Your product] vs [competitor]", on your own site
Why. Many vendors do not dare. They are wrong. The "Indy vs Pennylane" page on indy.fr is the 2nd most cited URL by ChatGPT across the whole category: 15 citations. The vendor who publishes the comparison becomes the one who frames the choice.
Which ones. The 3 competitors that show up with you in ChatGPT's answers. Not the ones you would like to beat. The ones the AI already compares you with.
The rule. Treat the competitor fairly and say where it is better. A hit piece is not picked up, because it contradicts the other sources the AI reads.
What goes in it:
| Block | Content |
|---|---|
| Title | [Product] vs [Competitor]: which one to choose in 2026? |
| Verdict in 3 lines | "Choose [Competitor] if… Choose [Product] if…" |
| Line-by-line table | Entry price, limits, key features, integrations, support |
| External proof | G2 or Capterra rating for both products, with the date |
| Screenshots | Both interfaces, with precise alt text |
| Last updated date | Visible at the top of the page |
Page 4: a "Best [category] software in 2026" post on your blog
Why. Of the 40 most cited URLs in the barometer, 23 are comparisons or rankings. Some vendors are cited as a source without even being on top: Dougs gets 11 citations from its blog for 3 recommendations, Sellsy 12 for 7. Their blog is raw material for the answer.
What goes in it. A real ranking of 5 to 10 tools, yours included, with visible criteria. A summary table at the top. A section titled with each tool's name. Each one's price. The year in the title.
Avoid. Ranking yourself first on every criterion. Rank yourself first on the use case where you really are (see page 5).
Page 5: one page per use case, in the buyer's words
Why. On "best invoicing software", the leader wins. On a precise question, the spot is open. Freebe shows up in only 6 answers out of 53. But on "freelancer, quotes, invoices and social contributions", it is first on all three Google runs. Abby doubles its share of citations on these concrete-need questions (from 8.6% to 17.2%), because its titles use the words those buyers use.
How to find the right pages. List what changes the recommendation in your category: team size, legal status, industry, budget, tool already in use. Then one page per combination that really changes the verdict.
What GPT-6 shows. On a CRM question for a small business, we saw ChatGPT ask its qualifying questions as buttons: team size, main need, budget per user. After the click, the list goes from 4 brands to 2, and sources move from review sites to vendors' official pages (support.pipedrive.com cited 12 times). It is a single conversation, but the message is clear: those buttons are the map of your use-case pages.
Example pages:
| Category | Example pages |
|---|---|
| CRM | CRM for a team of 2 to 5 sales reps · CRM for real estate agencies · CRM that connects to Gmail |
| Invoicing | Invoicing for freelancers · for agencies · for nonprofits |
| HR software | HR software for under 50 employees · for construction · with payroll included |
Page 6: the page on your market's hot topic
Why. Every category has its topic of the moment. In French invoicing, it is e-invoicing: 26 of ChatGPT's 72 searches. Tiime gets 14 citations with that single page. One well-made page, at the right time, can bring more than a series of average blog posts.
How to find yours. An incoming regulation (e-invoicing, AI Act, NIS2). A certification buyers require (SOC 2, ISO 27001, local hosting). An integration everyone asks for.
What goes in it. The yes / no answer in the first sentence ("[Product] has been a certified platform since…"). The dates. What the user has to do, step by step. The link to the official source.
Page 7: your spot in the comparisons the AI rereads
This page is not on your site. It is still often the most profitable one.
Why. Clubic's "5 best accounting software in 2026" ranks Pennylane, Indy, Tiime, Sage, EBP. ChatGPT recites that ranking in the same order (average ranks: 1.00, 2.00, 3.00, 3.78, 4.43). A single third-party page writes the answer.
Being listed is not enough. What counts is the top 3 and a section titled with your name. In RobZ IA audits, a brand ranked 5th in a top 10 did not make it into the answer. The same brand, with a section under its name on another page of the same site, came out first.
The warning sign. EBP is recommended 13 times in the barometer. Its site is cited 0 times. Others describe its product and its prices. Compute your ratio: citations of your site divided by your recommendations. Indy is at 1.79. Below 0.5, start with pages 1, 2 and 3.
How to do it:
- Ask your 10 main questions to ChatGPT and Google, three times each.
- Write down the URLs that come back on every run. Usually 5 to 8 pages per category.
- For each one: are you there? At what rank? Is the information right?
- Write to the publisher with a ready kit: current prices, screenshots, logo, one positioning sentence. These sites are often small and they reply.
Google bonus: nobody owns video yet
ChatGPT and Google AI Overviews do not read the same sources. YouTube gets 36 citations on Google and 0 on ChatGPT. Those 36 citations are spread over 20 videos. The best placed one gets only 4.
If you are a challenger, this is your way in. A 5-minute video, titled with the buyer's exact question ("Which invoicing software for a freelancer in 2026?"), with chapters and a transcript.
The checklist for every page
On the most cited pages of the barometer, six elements almost always come back:
| Element | Present on the most cited pages |
|---|---|
| The year in the title | 5 of 6 |
| A visible last updated date | 5 of 6 |
| An HTML table (not an image) | 5 of 6 |
| The answer in the first 100 words | 5 of 6 |
| An FAQ of 4 to 8 questions | 5 of 6 |
| Prices in numbers | 4 of 6 |
How to know it works
Write 20 questions your prospects really ask, half of them with a constraint. Ask each question at least three times per assistant. From one run to the next, the brands named barely move (stability index of 0.73 to 0.85 out of 1). Sources move a lot (0.45). A single run would send you to work on the wrong pages.
Measure again every two weeks and note what you published in between. Count 6 to 12 weeks before citations move on a new page.
Example questions for a CRM:
- Which CRM for a 5-person small business under $30 per user?
- HubSpot or Pipedrive for a small sales team?
- Best free CRM in 2026?
- A simple CRM that connects to Gmail?
Sources
RobZ IA GEO barometer, French invoicing and accounting software, August 2026: 9 buyer questions, 3 runs, ChatGPT and Google AI Overviews, 53 answers, 467 source citations, 72 ChatGPT searches recorded. See the barometer (in French).
GPT-6 observation: one ChatGPT conversation about choosing a CRM for a small business, 9 October 2026. Not reproduced over several runs.
Questions
- Which page should a SaaS start with in GEO?
- The pricing page. It is what ChatGPT checks first on a vendor's site. Put prices in HTML text, with limits and the date.
- Should you really publish comparisons with your competitors?
- Yes. A "us vs competitor" comparison published by a vendor is the 2nd most cited URL in the SaaS barometer. It has to stay fair to be picked up.
- How many use-case pages should you create?
- One per combination that changes the recommendation: team size, legal status, industry, budget. The qualifying questions ChatGPT asks in your category give you the list.
- Does GEO work the same on ChatGPT and Google?
- No. ChatGPT mostly cites vendor sites and official sources. Google cites more review sites and YouTube. Work on both.
- How long before you get cited?
- Count 6 to 12 weeks after a page goes live, when the AI already knows your brand. Longer if your brand appears in no source at all.
Read next
See where you stand today
Paste your site. We ask AI the questions your buyers type and count who gets named.