Periscopy
Articles
Methodology9 min read

Why your B2B SaaS does not appear in ChatGPT

Five technical reasons your B2B SaaS company is invisible in ChatGPT, Claude and Perplexity, and what to do today to turn it around.

Most Portuguese B2B SaaS companies are invisible in ChatGPT, Claude and Perplexity for five concrete technical reasons: no structured schema.org, no llms.txt, content with no citable structure, weak external authority, and low SEO ranking on the key queries. Fixing the first three is two to four weeks of work; the other two take three to six months.

Key takeaways
  • presence in AI engines is not a direct consequence of classic SEO
  • five technical causes explain 90% of invisibility cases
  • the first 30 days should focus on technical foundations (schema, llms.txt)
  • results in Perplexity appear in weeks; in ChatGPT, Claude and Gemini, in months. No deadline promised
  • whoever promises immediate results in GEO is selling an illusion

The problem, in numbers

A G2 survey (The Answer Economy, 2026) puts at 51% the share of B2B buyers who turn to an AI assistant more often than to Google, up from 29% eleven months earlier. That is the trajectory; nobody knows next year's number, and whoever states it is guessing.

For your company, this means half your prospects now start the comparison in a conversational interface, not in Google. If ChatGPT does not mention you when somebody asks for the best platform in your sector in Europe, you never reach the shortlist.

Most Portuguese B2B SaaS companies discover this by accident: somebody on the team asks ChatGPT, sees the competitors mentioned, and their own brand absent. The typical reaction is confusion. "But we appear in Google, don't we?" Yes. And it is not enough.

The five technical reasons

1. No structured schema.org

Schema.org is the vocabulary that describes, in machine-readable code, what each page is. Language models read schema before they read human text. Without schema, the site is literally less legible to the AI than it is to a human.

Most Portuguese B2B SaaS sites have minimal schema (a simple Organization, perhaps), but fail in three areas:

  • Service schema for each service offered
  • FAQPage schema on pages with questions and answers
  • Article schema with author, datePublished and dateModified on every blog article

Action: audit the current schema at validator.schema.org. If only Organization shows up and nothing else, there is work to do.

2. No llms.txt

llms.txt is a file at the root of the site, in Markdown, that describes the company in language optimised for models. It is still an emerging convention, but Anthropic, Mistral and several other large companies have adopted it.

Without llms.txt, AI crawlers have to guess your site's structure through HTML. With a well-made llms.txt, they get a curated summary, with a clear hierarchy of your most important pages.

Action: create /llms.txt at the root. Basic structure:

# Company name
> Description in one sentence

## Services
- [Service 1](URL)
- [Service 2](URL)

## Resources
- [Blog](URL)
- [Case studies](URL)

3. Content with no citable structure

Language models prefer to cite content with specific characteristics: clear definitions in the first 100 words, structured lists, explicit FAQs, hierarchical headers.

Your current blog articles probably follow the human logic of classic SEO: emotional hook, build-up of context, main point halfway through. That structure is terrible for AI citation. The models extract badly and cite little.

Action: rewrite your 5 to 10 most relevant articles with this structure: citable summary, key takeaways, structured body, FAQ, sources.

4. Weak external authority

The models do not cite only on the basis of your site's content, they also cite on the basis of how many credible external sources reference you. Wikipedia, credible software comparison sites (G2, Capterra), sector publications, academic papers: those are sources the models weigh heavily.

If your company is not on G2 with recent reviews, has no entry (or a poor entry) on Wikipedia, and is not mentioned in Portuguese and European sector publications, the external signal is missing.

Action: audit your presence on G2, Capterra and Wikipedia. Identify 3 to 5 relevant sector publications. Build a sustained editorial relationship, not a one-shot.

5. Low SEO ranking on the key queries

Counter-intuitive as it may seem, classic SEO still matters for GEO. Models with active web search (ChatGPT with browsing, Perplexity, Gemini with Google) go to Google's top 10 for sources to answer with. If you are not in the top 10 on the key queries, you will hardly be cited.

Action: identify the 20 queries that matter for your brand. Check your current position. If you are outside the top 10, there is classic SEO work to do before GEO bears fruit.

In what order to do it

If you start today, this is the order that works:

  1. Weeks 1 and 2: expanded schema.org, llms.txt, ai.txt, explicit robots.txt.
  2. Weeks 3 and 4: rewrite 5 articles with a citable structure.
  3. Month 2: start building external authority (G2, publications).
  4. Months 3 to 6: classic SEO in parallel, new articles, continuous monitoring.

Results in Perplexity start showing up between week 6 and week 10. In ChatGPT, Claude and Gemini, between month 3 and month 6.

Frequently asked questions

How long does it take to see results in GEO?

Perplexity: weeks (live web search). ChatGPT, Claude, Gemini: months (they depend on periodic training cuts and on accumulated authority). No deadline to promise; there is weekly measurement, and the before and after of each action.

Can I do GEO without doing SEO?

No. Models with web search go to Google's index. Without reasonable SEO, GEO yields little.

How much does doing this properly cost?

It depends. A technical diagnosis costs EUR 1,500 to EUR 3,000 in the European market. Continuous implementation runs between EUR 2,000 and EUR 8,000 a month depending on scale.

Can my in-house team do this?

Yes, if you have at least one technical person and one content person, with 5 to 10 hours a week allocated. The question is the learning time, and whether it is worth it in that equation.

Does ChatGPT update data in real time?

No. The version without web search answers from periodic training (generally months). The version with web search answers from the Bing index.

Which engine is most important to optimise first?

Perplexity, for two reasons: permanent web search (fast results) and a growing share of B2B searches.

Sources

Read next