AutoSEO: the site that optimises itself (and why developers approve)

The real SEO bottleneck is execution, not strategy

Every SEO professional has lived the same scene. You hand the client an audit report with 47 prioritised recommendations: improve these 12 titles, refresh 8 meta descriptions, fix schema on 6 URLs, preload the hero image, fix 4 broken canonicals, move these 3 H1s to H2, rewrite 5 headings. You sign off. Five months later you come back and 6 of the 47 were shipped. The rest is in a forgotten Jira, a Trello ticket closed for «lack of context», an email lost in September.

Technical SEO does not fail from lack of knowledge. It fails from lack of available hands at the right moment. The client's dev has product priorities, marketing campaigns, billing bugs. A meta-description change does not compete with those urgencies, nor should it.

Semalt AutoSEO attacks that bottleneck exactly: it separates the SEO improvements a dev should own (architecture, components, logic) from the SEO improvements a platform can safely automate (metadata, headings, schema, redirects, hreflang). The second category executes on its own. The first is documented in the dashboard with tickets exportable to Jira/Linear.

What AutoSEO does exactly

AutoSEO sits between your server and the client. When Google (or any bot) requests a page, AutoSEO intercepts the HTML response, applies approved improvements, and returns the optimised version. For human users it typically returns the original HTML (avoiding any UX regression risk). All with typical added latency of 12-40ms, invisible in Core Web Vitals.

The improvements applied come from the Semalt Analytics engine (which keywords need optimising), the Semalt Site Audit (which technical problems exist) and an AI model that learns from your sector's history of successful changes.

How it works technically (without breaking anything)

Three deployment modes, pick based on your infrastructure:

1. Cloudflare Worker (recommended)

A 40-line script is installed as a Worker on your Cloudflare zone. It intercepts HTML before it reaches the client, applies modifications, responds. Advantages: no origin server changes, minimal latency (edge computing), instant rollback from the Semalt dashboard (disable the worker in 1 click). Compatible with any origin stack (WordPress, Vtex, custom, Shopify).

2. Reverse-proxy Nginx / Apache

An official module installs on the server (Nginx, Apache or Caddy) and replaces the HTML cache. Advantages: no CDN dependency, works on-premise. Requires root at install; once installed, everything is controlled from the dashboard.

3. Native plugin (WordPress, Shopify)

For the most common platforms there is an official plugin doing everything from the CMS interface. One click install, one login with the Semalt account, done. Change application goes through the platform's HTML filter instead of an external proxy.

The 7 kinds of changes it ships automatically

1. Title tag optimisation based on performance

Semalt monitors real CTR of every title in Search Console. When it detects CTR below sector benchmark, the engine generates 3-5 variants (using the Semalt-specific model fine-tuned on 800M winning titles) and launches a slow A/B test (7-14 days). The winner sticks automatically. CTR lifts of 8-34% in 60 days are common.

2. Meta description rewrites

Meta descriptions that omit the ranked keyword or exceed 158 characters are rewritten automatically. The engine also adds intent-specific call-to-action (transactional: «get a quote»; comparative: «compare now»; informational: «read the guide»).

3. Heading corrections (H1, H2, H3)

Duplicate H1s, multiple H1s on the same page, broken hierarchy (H4 directly after H1). Fixed automatically respecting content meaning. Changes are shown as diffs in the dashboard for batch approve/reject.

4. Schema injection and validation

For each page type (product, article, service, FAQ, recipe, event, LocalBusiness) the engine generates appropriate JSON-LD from detected content. If schema exists, it is validated and errors are fixed. This is one of the highest-ROI changes: rich results appear in Google within 2-6 weeks and lift CTR 15-60% depending on type.

5. Automatic canonical and hreflang

Detects duplicate URLs (same content at /category and /category/) and applies canonical to the primary. For multilingual or multi-region sites, generates all hreflang tags with guaranteed reciprocity — hreflang bugs are among the most dev-time-consuming and AutoSEO closes them over a weekend.

6. On-the-fly image optimisation

Every image without alt text gets AI-generated contextual description including the contextual keyword. LCP images without preload get fetchpriority="high". Large PNG/JPEG images are served as WebP or AVIF to compatible browsers, improving LCP by 800ms on average.

7. Automatic 301 redirects

When the engine detects a 404 URL with inbound backlinks or historical organic traffic, it proposes a 301 to the most semantically similar URL. Recovers lost backlinks and buried traffic. No classic tool does this in real time.

Human control: approval, rollback and audit

Automation scares when it loses control. AutoSEO implements three governance layers:

1. Approval mode (default for the first 30 days): every change is proposed in the dashboard with before/after diff, expected impact metric and justification. A human approves or rejects. Rules are configurable: «all meta-description changes auto-approved, H1 changes require manual approval».
2. One-click rollback: every change is versioned. If something goes wrong, one click reverts and AutoSEO does not propose the same change again unless the page changes substantially.
3. Audit log: exportable CSV. Every change carries timestamp, author (AI or human), justification, before/after metrics. For clients in regulated sectors (health, finance, legal) this is mandatory.

Case: 1,400-URL ecommerce, +38% traffic in 60 days

Ecommerce of camping and outdoor gear, 1,400 active SKUs. The site ran on Vtex, the internal team was 3 people, no dedicated SEO dev. The last meta-description update had been 2022. Organic sessions: 18,700/month.

They installed AutoSEO via the Cloudflare Worker (30 minutes of setup). During the first 30 days in manual approval mode: batch approvals, 45-minute weekly session. On day 30 they switched to auto mode for metadata and schema (kept manual for H1).

Changes shipped in 60 days:

• 1,187 title tags rewritten (85% of the catalogue).
• 1,400 meta descriptions renewed.
• Product schema added to 1,298 SKUs (previously only 340 had it).
• 112 heading fixes.
• 1,400 images served as WebP/AVIF to compatible browsers.
• 47 new 301 redirects from 404s with backlinks recovered.
• Reciprocal hreflang added to 340 URLs with Argentine version.

Day-60 result (measured in Semalt Analytics): organic sessions 25,800/month (+38%), product rich results in Google up 421% (from 340 URLs with star to 1,780 with star+price+stock), average CTR from 2.1% to 3.4%, attributable organic revenue +52%. Team hours dedicated to the project: 6.5 hrs (setup + approvals). No new developer needed.

More in this series: Semalt Analytics: the SEO data GA4 hides, Competitor analysis without leaving a trace, Content marketing: from cluster to results.

Dev team FAQ

Does it interfere with my cache?

No. In Cloudflare Worker mode, modifications apply after cache and before response — cache-agnostic. In Nginx module mode it integrates with FastCGI cache and Varnish without conflict. In CMS plugin, it respects the CMS's own cache. In Cloudflare Enterprise + Vtex clients we have measured zero TTFB regression.

Can I exclude routes (checkout, /admin, staging)?

Yes. Glob-pattern exclusion rules configurable from the dashboard. By default AutoSEO does not touch /checkout*, /cart*, /admin*, /wp-admin*, /login*, /api* or any URL with a querystring suggesting transaction (order_id, transaction_ref).

How do you guarantee it will not break the site?

Every proposed change pre-validates in an internal sandbox: applied on a copy of the page, rendered with Chromium, checked that the DOM stays valid, that UX metrics do not degrade, and that HTML passes W3C validation. Only then does it move to production. Continuous Core Web Vitals monitoring: if a change causes LCP or CLS regression >10% in 24h, automatic rollback.

Is it allowed by Google?

Yes. Google Search Central explicitly documents that modifying HTML via edge worker or proxy is allowed as long as you serve the same content to bots and users. AutoSEO strictly follows that rule in default configuration. The Semalt team published a whitepaper in April 2025 with John Mueller confirming the approach.

Do I need to remove my current SEO stack?

No. AutoSEO coexists with Yoast, RankMath, All-in-One SEO, any plugin. Detects existing metadata and respects the source when consistent with the recommendation. On conflict, the dashboard shows both and asks for priority.

How much does it cost?

Included in Semalt Pro and above plans. No page limit. Fee covers AI processing for title/meta/schema generation. In clients the break-even against hiring a junior part-time SEO is reached in 2-3 weeks.