How to Write SEO Meta Titles and Descriptions at Scale (2026)

Writing one good meta title is easy. Writing 400 unique ones that match search intent, fit pixel limits, stay on brand, and actually earn clicks is a production problem. Most teams solve it by copying a template across every page and hoping Google does not notice. Google notices.

The short answer

Scale meta tags with a six-step loop: inventory every URL and label its page type, label intent (know, do, compare, buy) per page, generate two to three title and description variants with constrained AI prompts, score each variant against a rubric covering click promise, specificity, length, uniqueness, brand fit, and accuracy, review by a human with tiered depth (100 percent for money pages, sampling for long tail), then ship through your CMS and monitor CTR plus rewrite rates in Google Search Console.

Ship threshold is a rubric score of 80 or higher. No variant ships without a human check on P0 pages (home, pricing, top product). Long-tail pages can ship on a spot-check sample. The workflow works with any LLM and any CMS. The ILAA SEO Meta Generator handles generation for free with no sign-up, and pairs with the Grammar Checker for a polish pass and the Site AEO Audit tool for post-ship citability scoring.

Why scale breaks (and what "good" means in 2026)

Three things break when you push meta tags at volume:

1. Duplicate templates. Every product page gets "Product Name | Brand" and every blog post gets "Blog Post Title | Brand." Google sees 200 pages with structurally identical titles and treats them as low-effort. Pages that should rank for different queries start competing with each other.
2. Clickbait mismatch. AI tools told to "make it catchy" produce titles that overpromise. The page ranks, the searcher clicks, the content does not match, and the bounce tells Google the title was wrong. CTR drops, rankings follow.
3. Google rewrites. Google does not always use the title and description you write. It generates its own from page content and query context. If your tags are generic, Google replaces them with something you cannot control. If your tags are specific and accurate, Google keeps them more often.

"Good" in 2026 means the title and description accurately represent the page, match the searcher's intent, fit the pixel window on desktop and mobile, use the brand name consistently, and are unique across the site. Google's own guidance is straightforward: make titles descriptive and concise, make descriptions accurate and unique, and let Google know what you want it to display by writing good snippets rather than hoping it guesses.

Length, pixels, and rewrite reality

Google does not enforce hard character limits on titles or descriptions. It truncates based on pixel width, which varies by font, device, and query. The practical targets that work across most SERPs:

These are guidelines, not limits. Google's documentation on title links says titles should be descriptive and concise, and Google may use your title or generate one from page content and anchor text. Industry analysis from Search Engine Land notes that average displayed titles in SERPs often run 42 to 46 characters, shorter than the 60 most SEOs target, because Google truncates aggressively on mobile.

Moz recommends 50 to 60 characters for titles and under 160 characters for descriptions. Both are consistent with Google's pixel-based truncation behavior. The takeaway: front-load your primary keyword and the most compelling part of your value proposition. If Google truncates, the first 40 characters matter most.

The BATCH, SCORE, SHIP system

This is the operating system. One person can run it for a 50-page site. A team of three can run it for 5,000 pages. The steps are the same.

The system is deliberately boring. No fancy tooling required. A spreadsheet, an LLM, and a CMS. The discipline is in the rubric and the gates.

Step 1: Inventory export and page typing

Before generating anything, export a complete list of URLs you want to write or fix metas for. Pull these from your sitemap, Google Search Console, or your CMS export. For each URL, capture:

• URL (full path)
• Page type (homepage, product, pricing, blog post, category, docs, landing, legal, other)
• H1 (the on-page heading)
• Primary query (the one query this page should rank for)
• Brand token (how your brand appears in titles, e.g., "Acme" or "Acme CRM")
• Must-include facts (price, year, key feature, location, or other fact that must appear)
• Traffic tier (P0 for money pages, P1 for top blog or category, P2 for long tail)

Traffic tiers drive everything downstream. P0 pages get full human review. P2 pages can ship on a rubric score alone with spot-check sampling. If you have 500 URLs and 12 are P0, you review 12 by hand. That is manageable.

If your CMS does not export this data cleanly, use ILAA's Web Content Scraper to pull H1s and page text from live URLs, then assemble the spreadsheet manually. It is tedious for the first 50 rows and fast after that.

Step 2: Intent and SERP job per URL

Every page has a job in the SERP. A pricing page answers "how much does this cost?" A blog post answers "how do I do X?" A category page answers "what are my options?" If your title does not match the job, the searcher does not click.

Label each URL with one of four intents:

This label drives the generation prompt. A "know" page title should be descriptive and authoritative. A "buy" page title should be specific and include the brand. Do not mix patterns. A pricing page titled "Our Thoughts on Pricing" will not rank for "Acme pricing" because the title does not match the intent.

Step 3: Constrained batch generation

This is where AI does the heavy lifting. The goal is two to three variants per URL, generated with hard constraints. The mistake most teams make is pasting a full site's content into ChatGPT and asking for "good titles." You get generic output because the prompt has no constraints.

The generation prompt must include:

1. Brand voice rules: tone, forbidden phrases, brand token format.
2. Length constraints: title 50 to 60 characters, description 150 to 160 characters.
3. Must-include facts: per-page facts that must appear (price, year, feature, location).
4. Intent label: the SERP job for this URL.
5. Competitor context (optional): one or two competitor titles for tone calibration.
6. Facts vault: a one-sentence factual summary of what the page actually offers, so the AI does not invent features.

Here is the batch prompt block you can copy and run:

You are a metadata writer for {BRAND_NAME}.

Brand voice rules:
- Tone: {professional | casual | authoritative | friendly}
- Brand token: "{BRAND_TOKEN}" placed at the end of titles, separated by " | "
- Forbidden phrases: {list any banned words or phrases}
- Never use clickbait, hype words, or exclamation marks in titles
- Never fabricate features, prices, or facts not provided

Length constraints:
- Title: 50 to 60 characters including brand token
- Description: 150 to 160 characters
- Front-load the primary keyword in the title within the first 40 characters
- Place the most important sentence in the description first

For each URL below, generate 3 title variants and 3 description variants.

Page URL: {URL}
Page type: {homepage | blog | product | pricing | category | docs}
H1: {H1}
Primary query: {PRIMARY_QUERY}
Search intent: {know | do | compare | buy}
Must-include facts: {e.g., "Free plan available. Pro starts at $29/mo. SOC 2 in progress."}
Facts vault: {one sentence summary of what the page offers}

Return format:
TITLE_V1: ...
TITLE_V2: ...
TITLE_V3: ...
DESC_V1: ...
DESC_V2: ...
DESC_V3: ...
CHAR_COUNT_TITLE_V1: ##
CHAR_COUNT_DESC_V1: ###

Run this for batches of 10 to 20 URLs at a time. Larger batches degrade quality because the model starts repeating patterns. If you are using the ILAA SEO Meta Generator, it enforces length and brand constraints for you and returns ready-to-score variants without the prompt engineering overhead.

For teams using ChatGPT, Claude, or Gemini directly, the prompt above works with any current 2026 model. Do not reference deprecated model names or versions in your workflow documentation. Use "your LLM" or name the product by its current label.

Step 4: The Meta Quality Rubric

Every generated variant gets scored 0 to 100 across six criteria. This is the gate. No variant ships below 80.

Ship threshold: 80 total points. Score below 80 means the variant goes back for regeneration or human editing.

The rubric is intentionally simple enough to run in a spreadsheet. Add a column for each criterion, enter the score, sum the total. A content operator can score 50 variants per hour once they have the rubric memorized.

Step 5: Human quality gates by tier

AI generates. Humans decide. The depth of human review depends on the page's traffic and revenue importance.

The gate is not optional. AI-generated metas that ship without any human review will include duplicates, factual errors, and tone mismatches. The rubric catches most problems, but a human catches the rest. Budget one hour per 50 P0 variants and 30 minutes per 50 P1 variants.

If you are a solo founder or a one-person SEO team, this is where you spend your time. Generation is fast. Scoring is fast. The review is where you prevent damage.

Step 6: Rollout, CMS patterns, and monitoring

Once variants pass the rubric and human QA, deploy through your CMS. The deployment pattern depends on your stack:

WordPress (Yoast or RankMath)

Export your spreadsheet as CSV. Use the Yoast or RankMath bulk editor to paste titles and descriptions into the fields. Both plugins enforce character counts in their UI and show a SERP preview. Do a staged deploy: push P0 pages first, check live SERP snippets after three to five days, then push P1 and P2 in batches.

Webflow

Webflow's CMS lets you edit meta title and description fields per collection item. For large sites, use the Webflow API to batch update collection items. Export your final spreadsheet as JSON and PATCH each item.

Next.js or static site

Update frontmatter or the meta export in your page components. If you use a headless CMS like Sanity or Contentful, update the fields via their APIs. Commit changes, deploy, and verify with a curl check that the title and description tags render in the HTML.

Monitoring after deploy

Three things to watch in Google Search Console after shipping:

1. CTR by page (Performance report). Compare the 14-day period before and after the change. P0 pages should show CTR improvement within one to two weeks. If CTR drops, the new title or description is worse. Revert and regenerate.
2. Query coverage. Check whether new queries appear in the Performance report. Better titles often surface the page for query variations you were not ranking for before.
3. Actual SERP snippets. Search for your primary queries in an incognito window and compare the displayed title and description to what you wrote. If Google rewrites, note the pattern. If it consistently rewrites a specific page type, the title or description may not match the query intent closely enough.

Google's SEO starter guide reminds site owners that titles should be descriptive, unique per page, and brief. The monitoring step tells you whether you achieved that in practice.

Worked examples: homepage, blog post, product page

Three real-world before/after pairs with rubric notes. These are fictional brands but realistic patterns based on common SERP problems.

Example 1: SaaS homepage

Before:

• Title: Home | Acme
• Description: Welcome to our website. We offer innovative solutions for businesses of all sizes.

Rubric score: 22/100. Click promise 5 (no promise). Specificity 0 (no concrete detail). Length fit 10 (title too short). Uniqueness 5 (could be any homepage). Brand fit 5 (brand present but nothing else). Accuracy 0 (page actually offers AI meeting notes, not "innovative solutions").

After:

• Title: AI Meeting Notes for Remote Teams | Acme
• Description: Turn call recordings into action items in minutes. Free plan for small teams. SOC 2 in progress. Start without a credit card.

Rubric score: 93/100. Click promise 25 (specific promise: meeting notes from recordings). Specificity 20 (free plan, SOC 2, no credit card). Length fit 14 (title 42 chars, slightly short but fine). Uniqueness 15. Brand fit 15. Accuracy 10 (all claims true on the page).

Example 2: Blog post

Before:

• Title: Our Thoughts on Remote Work Productivity Tips and Tricks 2024
• Description: In this article we discuss remote work productivity and share some tips and tricks that might help you and your team.

Rubric score: 31/100. Click promise 5 ("our thoughts" is not a promise). Specificity 5 (year is stale). Length fit 5 (title 63 chars, over target). Uniqueness 10. Brand fit 10. Accuracy 0 (stale year, vague claim).

After:

• Title: Remote Work Productivity: 12 Habits That Stick (2026)
• Description: Twelve habits remote teams actually keep: async updates, focus blocks, and meeting caps. Includes a one-page checklist.

Rubric score: 95/100. Click promise 25 (12 specific habits). Specificity 20 (async updates, focus blocks, meeting caps, checklist). Length fit 15 (title 52 chars, description 149 chars). Uniqueness 15. Brand fit 10 (no brand token needed on a blog post with site-level branding). Accuracy 10.

Example 3: Product pricing page

Before:

• Title: Pricing
• Description: See our pricing.

Rubric score: 15/100. Every criterion near zero. The title is one word. The description is four words. Neither tells the searcher anything.

After:

• Title: Acme Pricing: Free, Pro, and Team Plans
• Description: Compare Free, Pro ($29/mo), and Team plans. Unlimited viewers on Pro. Annual saves 20%. No per-seat surprise fees on Free.

Rubric score: 91/100. Click promise 25 (plan names and prices). Specificity 20 (three plan tiers, $29/mo, 20% annual savings, unlimited viewers). Length fit 14 (title 41 chars, slightly short). Uniqueness 15. Brand fit 15. Accuracy 10 (all facts match the pricing table).

The pattern across all three: replace vague language with specific facts. Replace template structure with intent-matched structure. Replace stale or missing details with current, verifiable ones.

Templates: spreadsheet columns and batch prompt pack

Copy these spreadsheet headers into Google Sheets or Excel. One row per URL.

url | page_type | h1 | primary_query | search_intent | brand_token | must_facts | traffic_tier | title_v1 | title_v2 | title_v3 | desc_v1 | desc_v2 | desc_v3 | title_len | desc_len | rubric_score | qa_status | final_title | final_desc | shipped_date | notes

How to use it:

1. Fill columns 1 through 7 manually from your inventory export.
2. Generate title_v1 through title_v3 and desc_v1 through desc_v3 using the batch prompt from Step 3 or the SEO Meta Generator.
3. Auto-calculate title_len and desc_len with =LEN() formulas.
4. Score each variant and enter the total in rubric_score.
5. Set qa_status to "pending", "reviewing", "approved", or "rejected".
6. Copy the approved variant into final_title and final_desc.
7. Record the shipped_date when the change goes live.

For a 200-page site, this spreadsheet is your entire project management tool. Filter by traffic_tier to sequence your work. Sort by rubric_score to find variants that need another pass. The spreadsheet is the system.

Tool stack for 2026 (ILAA-first, fair alternatives)

You do not need paid tools to run this workflow. Here is the default stack:

The ILAA tools are free with no sign-up. Use them for the generation and QA steps, then move to your CMS for deployment. If you already pay for an SEO suite like Semrush or Ahrefs, use it for competitive title research (what are the top-ranking pages for your target queries using as titles?) but do not let it replace the rubric. The rubric is your quality gate, not a tool.

For more free tools across writing, code, marketing, and data, see the full 20 free AI tools you can use without signing up list.

Common mistakes that make AI meta tags worse

• Free-form dumping. Pasting your entire site content into an LLM without constraints produces generic titles. Always use the batch prompt with per-page facts vault and length limits.
• Template cloning. Using the same title structure for every product page ("Product Name | Brand") creates duplicate signals. Vary the angle: feature-led, use-case-led, audience-led, or comparison-led.
• Clickbait mismatch. "The Ultimate Guide" on a 600-word post. "7 Secrets" on a listicle with 3 generic tips. The click is wasted. Google learns from the bounce.
• Ignoring mobile. Titles that look great at 60 characters on desktop get truncated at 45 on mobile. Check both.
• No monitoring. Shipping metas and never checking GSC means you do not know if Google rewrote them or if CTR improved. Always compare before and after.
• Duplicate descriptions. If you cannot write a unique description, leave the tag empty. Google will generate one from page content. A duplicate description is worse than no description.
• Shipping all at once. Pushing 500 new title tags in one deploy makes it impossible to isolate what helped or hurt. Stage by traffic tier.
• Forgetting brand voice. AI defaults to a neutral marketing tone. If your brand is witty, terse, or technical, say so in the prompt. The AI Rewriter workflow covers voice preservation in depth.

How this fits your broader SEO and AEO work

Meta titles and descriptions are the SERP layer. They get the click. But getting cited by AI search engines like ChatGPT, Perplexity, and Google AI Overviews depends on what is on the page itself. Once your metas are clean, check these next:

• AEO in 2026: What Actually Gets You Cited by AI Search for the citation framework.
• How to Write Direct Answer Blocks That AI Search Engines Quote for body content that gets extracted.
• FAQPage Schema in 2026: Worth It and How to Add It with AI for structured data that helps AI engines parse your pages.
• How to Create FAQ-Ready Direct Answers for Product Pages for product and pricing page structure.
• Site AEO Audit in 15 Minutes for a tool walkthrough that scores any page for AI search citability.
• The Best Free AI Writing Tools in 2026 for the full writing tool stack.

FAQ

How do you write meta titles and descriptions at scale?

Export a URL inventory with page type and search intent, generate two to three variants per URL using constrained AI prompts, score each variant against a rubric covering click promise, specificity, length, uniqueness, brand fit, and accuracy, then run tiered human QA before shipping through your CMS. Monitor CTR and rewrite rates in Google Search Console after deployment.

What is a good meta title length in 2026?

Aim for 50 to 60 characters as a practical target. Google does not enforce a hard character limit. It truncates based on pixel width, which varies by device and font. Front-loading your primary keyword and brand name helps ensure the most important text displays before truncation.

What is a good meta description length in 2026?

Target 150 to 160 characters for desktop display. On mobile, Google may truncate around 120 characters. There is no hard limit on the meta description tag itself, but keeping key messaging in the first 120 characters ensures it survives truncation on most devices.

How do you keep brand voice when generating meta tags with AI?

Build a facts vault and brand voice rules into every generation prompt. Include your brand token format, tone descriptors, forbidden phrases, and must-include facts per page. Score outputs against a brand fit criterion in your rubric and have a human review every P0 and P1 page before shipping.

Should every page have a unique meta description?

Yes. Duplicate meta descriptions across pages signal low effort to search engines and reduce click-through rate. If you cannot write a specific description for a page, leave the tag empty and let Google generate one from page content rather than duplicating a template.

Does Google always use the meta title and description I write?

No. Google may rewrite your title or description based on the query and page content. Industry studies suggest Google rewrites titles 50 to 70 percent of the time. Write accurate, specific tags to minimize negative rewrites, and monitor your actual SERP snippets in Google Search Console.

Ship better metas this week

The workflow is deliberately simple so you actually do it. Export your inventory. Label intent. Generate with constraints. Score against the rubric. Review by tier. Ship in stages. Monitor CTR. The hard part is not the tooling. It is the discipline of scoring every variant and reviewing every P0 page by hand.

Start with your top 20 pages. That covers most of your traffic and revenue risk. Use the SEO Meta Generator for free, paste the output into the spreadsheet, score it, and ship. Once the top 20 are clean, work through P1 and P2 in batches of 50.

Ready to start? Open the ILAA SEO Meta Generator. Free, no signup, no card. Paste a page URL or a brief, get title and description variants in seconds.

Tool details and Google documentation links checked August 2026. Re-verify before publishing.