diff --git a/esmvalcore/_citation.py b/esmvalcore/_citation.py
new file mode 100644
index 0000000000..b3ead8a600
--- /dev/null
+++ b/esmvalcore/_citation.py
@@ -0,0 +1,247 @@
+"""Citation module."""
+import logging
+import os
+import re
+import textwrap
+from functools import lru_cache
+
+import requests
+
+from ._config import DIAGNOSTICS_PATH
+
+logger = logging.getLogger(__name__)
+
+REFERENCES_PATH = DIAGNOSTICS_PATH / 'references'
+
+CMIP6_URL_STEM = 'https://cera-www.dkrz.de/WDCC/ui/cerasearch'
+
+# The technical overview paper should always be cited
+ESMVALTOOL_PAPER = (
+    "@article{righi20gmd,\n"
+    "\tdoi = {10.5194/gmd-13-1179-2020},\n"
+    "\turl = {https://doi.org/10.5194/gmd-13-1179-2020},\n"
+    "\tyear = {2020},\n"
+    "\tmonth = mar,\n"
+    "\tpublisher = {Copernicus {GmbH}},\n"
+    "\tvolume = {13},\n"
+    "\tnumber = {3},\n"
+    "\tpages = {1179--1199},\n"
+    "\tauthor = {Mattia Righi and Bouwe Andela and Veronika Eyring "
+    "and Axel Lauer and Valeriu Predoi and Manuel Schlund "
+    "and Javier Vegas-Regidor and Lisa Bock and Bj\"{o}rn Br\"{o}tz "
+    "and Lee de Mora and Faruk Diblen and Laura Dreyer "
+    "and Niels Drost and Paul Earnshaw and Birgit Hassler "
+    "and Nikolay Koldunov and Bill Little and Saskia Loosveldt Tomas "
+    "and Klaus Zimmermann},\n"
+    "\ttitle = {Earth System Model Evaluation Tool (ESMValTool) v2.0 "
+    "-- technical overview},\n"
+    "\tjournal = {Geoscientific Model Development}\n"
+    "}\n")
+
+
+def _write_citation_files(filename, provenance):
+    """
+    Write citation information provided by the recorded provenance.
+
+    Recipe and cmip6 data references are saved into one bibtex file.
+    cmip6 data references are provided by CMIP6 data citation service.
+    Each cmip6 data reference has a json link. In the case of internet
+    connection, cmip6 data references are saved into a bibtex file.
+    Also, cmip6 data reference links are saved into a text file.
+    """
+    product_name = os.path.splitext(filename)[0]
+
+    tags = set()
+    cmip6_json_urls = set()
+    cmip6_info_urls = set()
+    other_info = set()
+
+    for item in provenance.records:
+        # get cmip6 data citation info
+        cmip6_data = 'CMIP6' in item.get_attribute('attribute:mip_era')
+        if cmip6_data:
+            url_prefix = _make_url_prefix(item.attributes)
+            cmip6_info_urls.add(_make_info_url(url_prefix))
+            cmip6_json_urls.add(_make_json_url(url_prefix))
+
+        # get other citation info
+        references = item.get_attribute('attribute:references')
+        if not references:
+            # ESMValTool CMORization scripts use 'reference' (without final s)
+            references = item.get_attribute('attribute:reference')
+        if references:
+            if item.identifier.namespace.prefix == 'recipe':
+                # get recipe citation tags
+                tags.update(references)
+            elif item.get_attribute('attribute:script_file'):
+                # get diagnostics citation tags
+                tags.update(references)
+            elif not cmip6_data:
+                # get any other data citation tags, e.g. CMIP5
+                other_info.update(references)
+
+    _save_citation_bibtex(product_name, tags, cmip6_json_urls)
+    _save_citation_info_txt(product_name, cmip6_info_urls, other_info)
+
+
+def _save_citation_bibtex(product_name, tags, json_urls):
+    """Save the bibtex entries in a bibtex file."""
+    citation_entries = [ESMVALTOOL_PAPER]
+
+    # convert tags to bibtex entries
+    if tags:
+        entries = set()
+        for tag in _extract_tags(tags):
+            entries.add(_collect_bibtex_citation(tag))
+        citation_entries.extend(sorted(entries))
+
+    # convert json_urls to bibtex entries
+    entries = set()
+    for json_url in json_urls:
+        cmip_citation = _collect_cmip_citation(json_url)
+        if cmip_citation:
+            entries.add(cmip_citation)
+    citation_entries.extend(sorted(entries))
+
+    with open(f'{product_name}_citation.bibtex', 'w') as file:
+        file.write('\n'.join(citation_entries))
+
+
+def _save_citation_info_txt(product_name, info_urls, other_info):
+    """Save all data citation information in one text file."""
+    lines = []
+    # Save CMIP6 url_info
+    if info_urls:
+        lines.append(
+            "Follow the links below to find more information about CMIP6 data:"
+        )
+        lines.extend(f'- {url}' for url in sorted(info_urls))
+
+    # Save any references from the 'references' and 'reference' NetCDF global
+    # attributes.
+    if other_info:
+        if lines:
+            lines.append('')
+        lines.append("Additional data citation information was found, for "
+                     "which no entry is available in the bibtex file:")
+        lines.extend('- ' + str(t).replace('\n', ' ')
+                     for t in sorted(other_info))
+
+    if lines:
+        with open(f'{product_name}_data_citation_info.txt', 'w') as file:
+            file.write('\n'.join(lines) + '\n')
+
+
+def _extract_tags(tags):
+    """Extract tags.
+
+    Tags are recorded as a list of strings converted to a string in provenance.
+    For example, a single entry in the list `tags` could be the string
+    "['acknow_project', 'acknow_author']".
+    """
+    pattern = re.compile(r'\w+')
+    return set(pattern.findall(str(tags)))
+
+
+def _get_response(url):
+    """Return information from CMIP6 Data Citation service in json format."""
+    json_data = None
+    if url.lower().startswith('https'):
+        try:
+            response = requests.get(url)
+            if response.status_code == 200:
+                json_data = response.json()
+            else:
+                logger.warning('Error in the CMIP6 citation link: %s', url)
+        except IOError:
+            logger.info('No network connection, '
+                        'unable to retrieve CMIP6 citation information')
+    return json_data
+
+
+def _json_to_bibtex(data):
+    """Make a bibtex entry from CMIP6 Data Citation json data."""
+    url = 'url not found'
+    title = data.get('titles', ['title not found'])[0]
+    publisher = data.get('publisher', 'publisher not found')
+    year = data.get('publicationYear', 'publicationYear not found')
+    authors = 'creators not found'
+    doi = 'doi not found'
+
+    if 'creators' in data:
+        author_list = [
+            item.get('creatorName', '') for item in data['creators']
+        ]
+        authors = ' and '.join(author_list)
+        if not authors:
+            authors = 'creators not found'
+
+    if 'identifier' in data:
+        doi = data['identifier'].get('id', 'doi not found')
+        url = f'https://doi.org/{doi}'
+
+    bibtex_entry = textwrap.dedent(f"""
+        @misc{{{url},
+        \turl = {{{url}}},
+        \ttitle = {{{title}}},
+        \tpublisher = {{{publisher}}},
+        \tyear = {year},
+        \tauthor = {{{authors}}},
+        \tdoi = {{{doi}}},
+        }}
+        """).lstrip()
+    return bibtex_entry
+
+
+@lru_cache(maxsize=1024)
+def _collect_bibtex_citation(tag):
+    """Collect information from bibtex files."""
+    bibtex_file = REFERENCES_PATH / f'{tag}.bibtex'
+    if bibtex_file.is_file():
+        entry = bibtex_file.read_text()
+    else:
+        entry = ''
+        logger.warning(
+            "The reference file %s does not exist, citation information "
+            "incomplete.", bibtex_file)
+    return entry
+
+
+@lru_cache(maxsize=1024)
+def _collect_cmip_citation(json_url):
+    """Collect information from CMIP6 Data Citation Service."""
+    json_data = _get_response(json_url)
+    if json_data:
+        bibtex_entry = _json_to_bibtex(json_data)
+    else:
+        bibtex_entry = ''
+    return bibtex_entry
+
+
+def _make_url_prefix(attribute):
+    """Make url prefix based on CMIP6 Data Citation Service."""
+    # the order of keys is important
+    localpart = {
+        'mip_era': '',
+        'activity_id': '',
+        'institution_id': '',
+        'source_id': '',
+        'experiment_id': '',
+    }
+    for key, value in attribute:
+        if key.localpart in localpart:
+            localpart[key.localpart] = value
+    url_prefix = '.'.join(localpart.values())
+    return url_prefix
+
+
+def _make_json_url(url_prefix):
+    """Make json url based on CMIP6 Data Citation Service."""
+    json_url = f'{CMIP6_URL_STEM}/cerarest/exportcmip6?input={url_prefix}'
+    return json_url
+
+
+def _make_info_url(url_prefix):
+    """Make info url based on CMIP6 Data Citation Service."""
+    info_url = f'{CMIP6_URL_STEM}/cmip6?input={url_prefix}'
+    return info_url
diff --git a/esmvalcore/_provenance.py b/esmvalcore/_provenance.py
index d0c5352e2b..bf675fae0b 100644
--- a/esmvalcore/_provenance.py
+++ b/esmvalcore/_provenance.py
@@ -79,7 +79,7 @@ def get_recipe_provenance(documentation, filename):
     entity = provenance.entity(
         'recipe:{}'.format(filename), {
             'attribute:description': documentation.get('description', ''),
-            'attribute:references': ', '.join(
+            'attribute:references': str(
                 documentation.get('references', [])),
         })
 
diff --git a/esmvalcore/_task.py b/esmvalcore/_task.py
index fe5136c7ea..d348f80ad3 100644
--- a/esmvalcore/_task.py
+++ b/esmvalcore/_task.py
@@ -17,6 +17,7 @@
 import psutil
 import yaml
 
+from ._citation import _write_citation_files
 from ._config import DIAGNOSTICS_PATH, TAGS, replace_tags
 from ._provenance import TrackedFile, get_task_provenance
 
@@ -565,6 +566,7 @@ def _collect_provenance(self):
             product = TrackedFile(filename, attributes, ancestors)
             product.initialize_provenance(self.activity)
             product.save_provenance()
+            _write_citation_files(product.filename, product.provenance)
             self.products.add(product)
         logger.debug("Collecting provenance of task %s took %.1f seconds",
                      self.name,
diff --git a/tests/integration/test_citation.py b/tests/integration/test_citation.py
new file mode 100644
index 0000000000..50fb65ef02
--- /dev/null
+++ b/tests/integration/test_citation.py
@@ -0,0 +1,118 @@
+"""Test _citation.py."""
+import textwrap
+
+from prov.model import ProvDocument
+
+import esmvalcore
+from esmvalcore._citation import (CMIP6_URL_STEM, ESMVALTOOL_PAPER,
+                                  _write_citation_files)
+from esmvalcore._provenance import ESMVALTOOL_URI_PREFIX
+
+
+def test_references(tmp_path, monkeypatch):
+    """Test1: references are replaced with bibtex."""
+    # Create fake provenance
+    provenance = ProvDocument()
+    provenance.add_namespace('file', uri=ESMVALTOOL_URI_PREFIX + 'file')
+    provenance.add_namespace('attribute',
+                             uri=ESMVALTOOL_URI_PREFIX + 'attribute')
+    filename = str(tmp_path / 'output.nc')
+    attributes = {
+        'attribute:references': 'test_tag',
+        'attribute:script_file': 'diagnostics.py'
+    }
+    provenance.entity('file:' + filename, attributes)
+
+    # Create fake bibtex references tag file
+    references_path = tmp_path / 'references'
+    references_path.mkdir()
+    monkeypatch.setattr(esmvalcore._citation, 'REFERENCES_PATH',
+                        references_path)
+    fake_bibtex_file = references_path / 'test_tag.bibtex'
+    fake_bibtex = "Fake bibtex file content\n"
+    fake_bibtex_file.write_text(fake_bibtex)
+
+    _write_citation_files(filename, provenance)
+    citation_file = tmp_path / 'output_citation.bibtex'
+    citation = citation_file.read_text()
+    assert citation == '\n'.join([ESMVALTOOL_PAPER, fake_bibtex])
+
+
+def mock_get_response(url):
+    """Mock _get_response() function."""
+    json_data = False
+    if url.lower().startswith('https'):
+        json_data = {'titles': ['title is found']}
+    return json_data
+
+
+def test_cmip6_data_citation(tmp_path, monkeypatch):
+    """Test2: CMIP6 citation info is retrieved from ES-DOC."""
+    # Create fake provenance
+    provenance = ProvDocument()
+    provenance.add_namespace('file', uri=ESMVALTOOL_URI_PREFIX + 'file')
+    provenance.add_namespace('attribute',
+                             uri=ESMVALTOOL_URI_PREFIX + 'attribute')
+    attributes = {
+        'attribute:mip_era': 'CMIP6',
+        'attribute:activity_id': 'activity',
+        'attribute:institution_id': 'institution',
+        'attribute:source_id': 'source',
+        'attribute:experiment_id': 'experiment',
+    }
+    filename = str(tmp_path / 'output.nc')
+    provenance.entity('file:' + filename, attributes)
+
+    monkeypatch.setattr(esmvalcore._citation, '_get_response',
+                        mock_get_response)
+    _write_citation_files(filename, provenance)
+    citation_file = tmp_path / 'output_citation.bibtex'
+
+    # Create fake bibtex entry
+    url = 'url not found'
+    title = 'title is found'
+    publisher = 'publisher not found'
+    year = 'publicationYear not found'
+    authors = 'creators not found'
+    doi = 'doi not found'
+    fake_bibtex_entry = textwrap.dedent(f"""
+        @misc{{{url},
+        \turl = {{{url}}},
+        \ttitle = {{{title}}},
+        \tpublisher = {{{publisher}}},
+        \tyear = {year},
+        \tauthor = {{{authors}}},
+        \tdoi = {{{doi}}},
+        }}
+        """).lstrip()
+    assert citation_file.read_text() == '\n'.join(
+        [ESMVALTOOL_PAPER, fake_bibtex_entry])
+
+
+def test_cmip6_data_citation_url(tmp_path):
+    """Test3: CMIP6 info_url is retrieved from ES-DOC."""
+    # Create fake provenance
+    provenance = ProvDocument()
+    provenance.add_namespace('file', uri=ESMVALTOOL_URI_PREFIX + 'file')
+    provenance.add_namespace('attribute',
+                             uri=ESMVALTOOL_URI_PREFIX + 'attribute')
+    attributes = {
+        'attribute:mip_era': 'CMIP6',
+        'attribute:activity_id': 'activity',
+        'attribute:institution_id': 'institution',
+        'attribute:source_id': 'source',
+        'attribute:experiment_id': 'experiment',
+    }
+    filename = str(tmp_path / 'output.nc')
+    provenance.entity('file:' + filename, attributes)
+    _write_citation_files(filename, provenance)
+    citation_url = tmp_path / 'output_data_citation_info.txt'
+
+    # Create fake info url
+    fake_url_prefix = '.'.join(attributes.values())
+    text = '\n'.join([
+        "Follow the links below to find more information about CMIP6 data:",
+        f"- {CMIP6_URL_STEM}/cmip6?input={fake_url_prefix}",
+        '',
+    ])
+    assert citation_url.read_text() == text
diff --git a/tests/integration/test_recipe.py b/tests/integration/test_recipe.py
index c9efc7a467..e6f64b8ad0 100644
--- a/tests/integration/test_recipe.py
+++ b/tests/integration/test_recipe.py
@@ -1190,11 +1190,6 @@ def simulate_diagnostic_run(diagnostic_task):
             'name': 'Bouwe Andela',
         },
     },
-    'references': {
-        'acknow_author': "Please acknowledge the author(s).",
-        'contact_authors': "Please contact the author(s) ...",
-        'acknow_project': "Please acknowledge the project(s).",
-    },
     'projects': {
         'c3s-magic': 'C3S MAGIC project',
     },
@@ -1271,7 +1266,7 @@ def test_diagnostic_task_provenance(
                                             key).pop() == record[key]
 
     # Check that diagnostic script tags have been added
-    for key in ('statistics', 'domains', 'authors', 'references'):
+    for key in ('statistics', 'domains', 'authors'):
         assert product.attributes[key] == tuple(TAGS[key][k]
                                                 for k in record[key])
 
@@ -1287,7 +1282,7 @@ def test_diagnostic_task_provenance(
     for key in ('description', 'references'):
         value = src['documentation'][key]
         if key == 'references':
-            value = ', '.join(TAGS[key][k] for k in value)
+            value = str(src['documentation'][key])
         assert recipe_record[0].get_attribute('attribute:' +
                                               key).pop() == value