How Leverge
negotiates on
your behalf.

When you enter a publisher URL (like techcrunch.com), our AI performs multi-stage analysis:

• Publisher profile — who they are, their audience, their editorial style
• Pricing intelligence — estimated rate card with confidence bands, based on comparable publishers
• Leverage analysis — what advantages you have (timing, quality content, volume, relationships)
• Contact discovery — who to email, with verified addresses where available
• Recommended strategy — opening price, target price, walk-away price, and tactics to deploy

How the leverage score works

The score (0-10) quantifies your negotiating position across five dimensions:

• Content quality — is your material ready-to-publish or will they need editorial work?
• Timing — are you negotiating during their slow season (Q1, August)?
• Audience match — does your story genuinely fit their readers?
• Volume leverage — are you offering one placement or a bundle?
• Relationship signals — have you published with them before?

A score of 7+ means strong position — expect 30-45% discounts. Below 5, you're negotiating uphill, and the AI will advise a more conservative approach.

OAuth flow explained

When you click "Connect Gmail," you're redirected to Google's consent screen (not Leverge's). Google asks: "Leverge is requesting these permissions. Allow?"

You approve on Google's domain. Google returns an OAuth token to Leverge — this token lets us send email through Gmail's API on your behalf. We never see your Google password.

What we do NOT access

Our scope is narrowly defined:

• We do not read your full inbox
• We do not scan emails you've received from others
• We do not access your contacts, calendar, or drive
• We do not see email addresses you haven't explicitly added to Leverge

We only read emails in threads you initiated through Leverge. If we send an email to alex@publisher.com and they reply, we can see that reply. If someone else emails you unrelated to Leverge, we never see it.

Revoking access

Disconnect any time from Leverge's settings, or directly from your Google account at myaccount.google.com/permissions. Revocation is immediate.

Here's a real example negotiation, anonymized, showing the actual back-and-forth over 3 rounds:

The tactics the AI uses

The AI deploys specific negotiation tactics based on your leverage profile and the publisher's responses:

• Anchoring — open below market to establish a favorable range. Opens at 45-55% below list when leverage is high.
• Seasonal timing — reference slow periods to justify discounts.
• Quality framing — emphasize ready-to-publish content that saves their editorial team time.
• Credible walk-away — signal willingness to pursue alternatives. Never bluffs.
• Volume leverage — bundle placements when you have multiple campaigns.
• Splitting the difference — when positions are close, propose mid-point to close quickly.

How it decides what to say

Every AI response is generated with full context:

• The full conversation history — every email sent and received
• Your research intelligence — leverage score, pricing bands, recommended strategy
• Your campaign parameters — target price, ceiling, walk-away threshold
• The publisher's signals — are they being flexible, firm, or pushing back?
• The round number — early rounds stay exploratory, later rounds push to close

When the AI pauses automatically

• Round 3 checkpoint — midpoint review: "Here's what they're offering. Accept, counter, or walk?"
• Round 5 checkpoint — the AI will not continue past round 5 without your explicit direction
• Offer exceeds ceiling — if any offer goes above your walk-away price, the AI stops immediately
• Hostile language detected — if the publisher's tone turns aggressive, the AI hands off to you
• Unexpected request — if the publisher asks for something outside scope, you review

What you see at a checkpoint

• The full conversation history up to this point
• The current offer vs. your target and ceiling
• The AI's recommended response (fully drafted)
• The AI's reasoning — why it suggests this approach
• Four options: Approve & Send, Edit & Send, Counter, Walk Away

Manual intervention at any time

• Pause an entire campaign — freezes all ongoing negotiations
• Pause a specific thread — take manual control of one publisher
• Edit any draft — override whenever needed
• Reply directly — if you reply manually in Gmail, the AI sees it and continues from your point

Closing signal detection

The AI detects close by analyzing the publisher's language for explicit agreement — phrases like "Deal," "Agreed at [price]," or "Sending contract." Ambiguous replies ("Let me think about it") are not treated as acceptance.

Every deal close requires your final confirmation. The AI never commits you without your explicit approval.

Savings calculation

Savings are calculated as:

Savings = List Price - Agreed Price

The "list price" comes from the publisher's initial quote in the thread. If the publisher didn't quote a price, we use the mid-range from your Research Intel report. This is always shown transparently on your dashboard.

Post-deal handoff

After a deal closes, Leverge:

• Records the final price, savings, and full conversation history
• Marks the contact as "deal closed" to prevent duplicate outreach
• Updates your analytics dashboard in real-time
• Sends you a summary email with next steps

The actual content submission and invoicing happen outside Leverge — we negotiate the price, you handle the execution. Your relationship with the publisher continues directly.

Budget enforcement

When you set a ceiling price (say $2,500), the AI will never agree to anything above it. If the publisher won't go below your ceiling after all rounds, the AI walks away cleanly — no overcommitting, no "let me check with my team" stalling.

Round limits

By default, the AI runs a maximum of 5 rounds. This prevents endless back-and-forth that wastes everyone's time. You can increase this to 7 for complex deals, but the AI actively pushes toward resolution rather than extending conversations.

Tone guardrails

The AI is always professional and respectful. It will not:

• Use aggressive or threatening language
• Make false claims about competing offers
• Misrepresent your company or intentions
• Promise things you haven't authorized

If a publisher becomes hostile, the AI de-escalates and pauses for your review rather than matching tone.

AI model

Leverge uses Claude by Anthropic as its negotiation engine. We chose Claude for its strong reasoning capabilities and built-in safety alignment. The model processes your full conversation history, research data, and strategy parameters for each response.

Email integration

Emails are sent through the Gmail API and Microsoft Graph API (for Outlook). We use OAuth 2.0 with minimal scopes. Email content is encrypted in transit and at rest. OAuth tokens are encrypted with AES-256.

Data handling

All data is stored in PostgreSQL with encryption at rest. We don't use your negotiation data to train AI models. Your conversations are private to your account.

• Database: PostgreSQL on AWS (encrypted at rest)
• Email credentials: AES-256 encrypted, keys in AWS KMS
• Application: Next.js on Vercel, serverless functions
• Background jobs: BullMQ + Redis for async email processing