Draft Landing-Page SEO Fields with Airtable and OpenAI

This landing-page series starts by drafting title tags, descriptions, and headings in Airtable. AI can produce a first draft from approved page facts. You still decide what gets published.

Prepare your Pages table

Use Name, Approved Facts, Draft SEO Title, Draft Description, Draft H1, and Draft H2 text fields. Add a record ID input to an Airtable Run a script automation. Create secrets named openaiKey and an input named model for a currently available Responses-compatible model. Keep approved facts specific. Do not ask the model to invent testimonials, prices, or locations.

Generate and save one draft

This replacement example uses OpenAI's Responses API. It makes one request, checks the result, and writes draft fields to the triggering record.

const {recordId, model} = input.config();
const table = base.getTable('Pages');
const row = await table.selectRecordAsync(recordId);
if (!row || !model) throw new Error('Missing record or model');
const facts = row.getCellValueAsString('Approved Facts').trim();
if (!facts) throw new Error('Approved facts required');
const response = await fetch('https://api.openai.com/v1/responses', {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${input.secret.openaiKey}`,
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    model, store: false, max_output_tokens: 600,
    instructions: 'Use only the supplied facts. Return JSON only with title, description, h1, h2 string keys. No markdown. No invented claims.',
    input: JSON.stringify({name: row.getCellValueAsString('Name'), facts})
  })
});
if (!response.ok) throw new Error(`OpenAI failed: ${response.status}`);
const result = await response.json();
if (result.status !== 'completed') throw new Error('Incomplete generation');
const text = (result.output || []).flatMap(item => item.content || [])
  .filter(part => part.type === 'output_text').map(part => part.text).join('');
const draft = JSON.parse(text);
for (const key of ['title', 'description', 'h1', 'h2']) {
  if (typeof draft[key] !== 'string' || !draft[key].trim()) throw new Error(`Invalid ${key}`);
}
await table.updateRecordAsync(recordId, {
  'Draft SEO Title': draft.title.trim(),
  'Draft Description': draft.description.trim(),
  'Draft H1': draft.h1.trim(),
  'Draft H2': draft.h2.trim()
});
output.set('recordId', recordId);

updateRecordAsync() saves Airtable fields. output.set() only exposes a value to later automation actions. Airtable documents that distinction. Malformed output stops this example instead of sending partial copy to Webflow.

Review before syncing

Compare every draft with its source facts and page content. Keep titles distinct across landing pages. Remove repeated location text and unsupported superlatives. Add an Approved checkbox that only a reviewer changes. Keep generation separate from the approval trigger so a draft cannot approve itself.

For location pages, build the listing-page structure first. Then send approved fields to Webflow. The next part explains the series' update workflow, including Last CMS Update.

Watch the original lesson.