Yes, LLMs like ChatGPT, Claude or Perplexity can change tenders.
They do not only change the wording. They can change what is requested, what is omitted, what becomes mandatory and what remains optional.
In complex B2B buying, the critical moment often happens before the tender is published: the buyer uses ChatGPT, Claude, Gemini or Perplexity to understand the market, frame the need, prepare criteria, draft a first RFP version or challenge an internal proposal.
Short answer
LLMs change tenders at five levels:
| Level | What the LLM can do | Main risk |
|---|---|---|
| Framing | Reformulate the need and objectives | Simplifying the real problem |
| Criteria | Suggest a selection grid | Installing generic criteria |
| RFP | Draft technical requirements | Omitting context, threshold or proof |
| Comparison | Prepare supplier questions | Comparing non-equivalent offers |
| Rejection | Identify blocking points | Keeping weak offers admissible |
The real risk is not only that an LLM can be wrong. The real risk is that it produces a request that looks correct, but is too weak for the real context.
What buyers are really searching for
Searches around AI and tenders often mix several intents: writing a tender faster, structuring an RFP, comparing suppliers, identifying risks or preparing an evaluation grid.
For a supplier, the question is therefore not only whether ChatGPT writes the final document. The question is which criteria AI installs before the document is frozen.
| Search intent | Buyer goal | Supplier risk |
|---|---|---|
| Write a tender with AI | Save time on the first version | A generic specification becomes the working base |
| Build an RFP with ChatGPT | Turn a need into requirements | Usage context, thresholds and proof remain implicit |
| Compare suppliers with AI | Get a clear evaluation grid | Criteria visible from competitors become dominant |
| Prepare a supplier shortlist | Reduce uncertainty before consultation | Less documented providers leave the shortlist too early |
| Secure a technical purchase | Avoid a poor decision | Declarative compliance replaces verifiable proof |
The right answer is not to publish generic content about “AI in tenders”. It is to document the criteria, proof, thresholds and usage situations that the buyer should find when using an LLM.
What Beyond Mentions data shows
In the first consolidated wave of the Beyond Mentions Observatory, we analyzed 4,320 Perplexity sonar answers over 3 UTC days, with 6 passes per question per day.
These figures do not directly measure all tenders. They describe a controlled corpus of AI answers used to observe how generative engines use sources, criteria, proof and standards in B2B decision situations.
| Observed signal | Volume in the corpus | Reading for tenders |
|---|---|---|
| Source dependency | 4,137/4,320 (95.8%) | LLMs rely heavily on reusable sources |
| Proof reuse | 3,714/4,320 (86.0%) | Published proof can become reused requirements |
| Documentation and proof | 2,687/4,320 (62.2%) | Documentation structures the decision before supplier contact |
| Shortlist and vendor evaluation | 2,168/4,320 (50.2%) | AI answers often activate selection logic |
| Criteria reuse | 1,464/4,320 (33.9%) | Explicit criteria can be reused in the buying grid |
| Specification gap | 1,130/4,320 (26.2%) | Technical capabilities can be poorly formulated when scope stays implicit |
The signal to watch here is Specification gap: when context, proof or threshold is not documented, AI can help formulate a plausible but insufficient requirement.
How an LLM changes a tender
1. It intervenes before the final RFP
A buyer can ask:
- “Which criteria should we include in a tender for this type of solution?”
- “Which standards should we require to avoid a poor decision?”
- “Which proof should suppliers provide?”
- “Which questions should we ask during the presentation?”
- “Which suppliers should we shortlist?”
- “How should we draft an RFP for this need?”
These questions can influence the specification before suppliers ever see the project.
2. It turns available content into requirements
An LLM does not know your domain judgment if it has not been documented. It reconstructs an answer from available sources, categories, criteria and proof.
| If your documentation contains… | The LLM can reuse… |
|---|---|
| Named choice criteria | An evaluation grid |
| Standards tied to context | A contextualized technical requirement |
| Verifiable proof | A requested supplier document |
| Thresholds | A minimum condition |
| Usage limits | A scope clause |
| Rejection criteria | An exclusion rule |
Conversely, if these elements remain implicit, the LLM can keep the simplest consensus.
3. It can lower the requirement level without visible error
Underspecification is more subtle than hallucination.
| Plausible AI wording | What is missing | Risk |
|---|---|---|
| ”Require the applicable standard” | Exposure scenario | The minimum standard appears sufficient |
| ”Ask for customer references” | Comparable proof | Declarative proof looks equivalent |
| ”Compare features” | Risk criterion | Price becomes dominant again |
| ”Check compliance” | Validation method | The supplier can declare without proving |
| ”Evaluate support” | Operational threshold | Premium service remains vague |
An RFP can therefore look coherent and still be insufficient in reality.
The most dangerous case: the minimum standard becomes the requested standard
For a premium, technical or regulated offer, value often depends on a requirement level above the market minimum.
If the LLM does not see why that level is necessary, it can recommend:
- a correct but too-weak standard;
- generic proof;
- a feature-centered comparison;
- an overbroad shortlist;
- missing rejection criteria.
Result: robust offers become comparable with cheaper offers because the specification does not request what justifies the gap.
Diagnosis: is your market exposed?
| Situation | Risk level | Why |
|---|---|---|
| Buyers do not master the technical topic | High | They ask the LLM to reduce uncertainty |
| Compliance depends on context | High | The standard can change with exposure |
| Proof is hard to compare | High | The LLM may accept declarative claims |
| Price appears too early | Medium to high | Value criteria are not explicit enough |
| Tenders all look similar | Medium | The LLM may reinforce generic standards |
| Your advantage is mainly methodological | High | It must be translated into criteria and proof |
The more complex the purchase, the more LLM-assisted drafting can move value before the first sales conversation.
What an LLM-assisted RFP should contain
A stronger RFP should not only list requirements. It should connect each requirement to its decision context.
| Block | Question it answers |
|---|---|
| Usage situation | In which context will the solution be used? |
| Exposure level | Which frequency, criticality or constraint changes the standard? |
| Expected standard | Which level is sufficient, and when does it become insufficient? |
| Verifiable proof | Which document, test, case or audit can verify it? |
| Threshold | When does the answer become insufficient? |
| Usage limit | When should the offer not be selected? |
| Rejection criterion | Which absence should exclude an answer? |
This logic is detailed in the Specification Gap: the gap between what a buyer asks for and what a complex decision actually requires.
Example: weak wording and stronger wording
Weak wording
“The supplier must provide a solution compliant with market standards and provide relevant references.”
This sentence may look correct. It remains too weak for a complex decision: it specifies neither context, proof, threshold nor rejection criterion.
Stronger wording
“The supplier must connect the proposed standard to the exposure scenario, provide verifiable compliance proof, document usage limits and specify the cases where the offer should not be selected.”
The second wording helps an LLM produce a more useful requirement because it encodes the decision logic.
What to publish before the tender
| Asset to publish | Expected effect in LLM answers |
|---|---|
| Buyer criteria guide | Helps structure the evaluation grid |
| Standards / Exposure Matrix | Connects standard, context, proof and rejection |
| Pre-RFP FAQ | Answers questions asked before drafting |
| Risk scenarios | Shows the cost of the wrong standard |
| Extractable proof | Gives the LLM verifiable elements to reuse |
| Rejection criteria | Prevents weak offers from remaining admissible |
| Contextual comparison | Avoids comparison between non-equivalent offers |
The ideal format is simple: buyer question, context, criterion, proof, threshold, limit, rejection.
How to audit tender risk
Test the questions your buyers could ask an LLM:
- “How should we write an RFP for this type of solution?”
- “Which criteria should we compare to choose a supplier?”
- “Which proof should we request before signing?”
- “Which standards should we require in this context?”
- “Which risks should we check before the tender?”
- “Which criteria should exclude an offer?”
Then code the answers:
| Element to check | Warning signal |
|---|---|
| Context | The LLM discusses the need without a usage scenario |
| Standard | The minimum is presented as sufficient |
| Proof | Compliance remains declarative |
| Threshold | No critical level is formulated |
| Rejection | All offers seem admissible |
| Comparison | Price or features dominate |
If these signals come back, the problem is not only a content issue. It is a commercial underspecification risk.
Key takeaway
LLMs change tenders because they intervene when the buyer formulates the need.
They can help structure a request better, but they can also:
- lower the requirement level;
- omit critical proof;
- make premium offers comparable with standard offers;
- turn a technical advantage into a secondary option;
- keep too many offers admissible because rejection criteria are missing.
The strategic question is not:
“Will LLMs write tenders?”
The real question is:
“Which criteria do LLMs give your buyers before the tender is frozen?”
Read next
- Specification Gap: understand underspecification risk.
- GEO: what happens after brand citations?: connect visibility, criteria, proof and shortlist inclusion.
- How does AI build a supplier shortlist?: see how criteria influence selected suppliers.
- Documentation Blind Spot: identify missing criteria and proof in your content.
FAQ
Are LLMs already writing tenders?
They can help prepare a tender, structure a need, rewrite an RFP or generate a first criteria grid. The main risk is not only error, but underspecification.
What is the main risk for a premium supplier?
The risk is that the LLM keeps a minimum standard, omits expected proof or turns a technical advantage into a secondary option. The premium offer then becomes comparable with less robust alternatives.
How do you know if an AI-assisted tender is under-specified?
Check whether the RFP connects each requirement to an exposure context, verifiable proof, threshold, usage limit and rejection criterion.
What should be published to reduce this risk?
Publish standards/exposure matrices, expected proof, risk scenarios, rejection criteria and buyer FAQs that LLMs can reuse before the RFP is written.
What question does the buyer ask AI?
Which documentation simplification can lower the standard?
Which technical requirement must be clearly formulated?
Which evidence should be requested or published?
Which criterion excludes an insufficient answer?